How many have had #COVID19 in Delhi? What % of infections have been detected? What is the fatality rate?
A (longish) #thread on Delhi's epidemic, with some analysis of its three serosurveys + other data. Details in a technical document linked at the end. (1/11)
First, Delhi's current surge (which may be winding down) is real - not just about better detection. But the actual surge in infections has been considerably smaller than in June - detection has increased a lot, making it seem larger. (Similar story to Mumbai - more later.) 2/
Prevalence estimate: by mid-August between 37% and 49% of Delhi people had had COVID. By mid-September: between 43% and 60%. The wide range reflects many uncertainties. When the October serosurvey results are out, we'll know more. (bloombergquint.com/coronavirus-ou…) 3/
Delhi's three serosurveys gave antibody levels of:
- 23% (July)
- 29% (August)
- 25% (September)
Interpretating the surveys is hard because of:
- poor transparency: no technical documentation
- changes in methodology & test-kit between surveys
- no sampling correction
4/
Waning antibodies (i.e., decreasing sensitivity of the antibody tests) helps explain Delhi's seroprevalence data. Estimate: at least 20% of positive tests turn negative each month. Probably closer to 30%. This determines where the truth lies between... 5/ ndtv.com/india-news/aug…
...two extremes:
- rapidly waning test sensitivity. High % have had COVID. Low % of infections detected. Low IFR.
- slowly waning test sensitivity. Lower % have had COVID. Higher % of infections detected. Higher IFR.
(See the analysis for Mumbai.) 6/ science.thewire.in/health/mumbai-…
Detection of infections: to date, between 2.5% and 3.5% of Delhi's total COVID-19 infections have been detected. But at the moment it's higher: my estimate is about 8% of infections are currently being detected. (95% CI: 4-16%). Why is infection detection rising?... 7/
Why are more infections being detected? 1) There is more testing - even correcting for the high proportion (currently ~80%) which are rapid 2) *Maybe* spread has shifted towards middle-class areas where detection is higher? (As in Mumbai - below) 8/ science.thewire.in/health/mumbai-…
Fatality rate: Delhi's naive infection fatality rate (ignoring undercounting) is low:
- overall ~0.06% (95% CI: 0.05-0.07%).
- currently ~0.085% (95% CI: 0.04%-0.18%).
There is some indication that it has been rising - fourth serosurvey data will make this clearer. 9/
A mystery: why is Delhi's naive IFR so low (0.05-0.07%)? Even correcting for plausible levels of undercounting IFR is low. Possibly the answer lies in uneven spread: have the elderly been partly shielded because of slower spread in planned colonies? 10/ scroll.in/latest/974615/…
The approach (Monte Carlo) is to put plausible distributions on some of the many unknown quantities and run lots of experiments (10,000) to get distributions on prevalence, IFR, detection of infections, etc. (11/11)
The documented instances of COVID-19 death undercounting are too many and too major in scale to be treated as aberrations. Many have followed similar patterns - for example omitting deaths from "comorbidities" and suspected deaths.
The "urban areas have high MCCD coverage" argument is something of a red herring. Delhi, with its high MCCD coverage, saw *huge* undercounting before subjected to pressure. Mumbai too added in lots of missed COVID-19 deaths in June (~1700).
I took a look at Mumbai's #COVID19 data by age and came to some worrying conclusions. I'd encourage people celebrating seemingly low IFR values to look more closely. I used: Spain's IFR data; and Mumbai's age pyramid, age-structured fatality data, and seroprevalence data. 1/6
Naive IFR from Mumbai's seroprevalence and fatality data is ~0.12% (not, I think, 0.05-0.1% as is widely quoted).
An "expected IFR" for Mumbai, using Mumbai's age pyramid and Spain's age-dependent IFR is ~0.22%.
These two values hide a more complicated story. 2/6
IFR in some age groups, particularly 40-60, appears to be quite a lot higher (about 60% higher) in Mumbai than in Spain, even *assuming no death undercounting*. If there is death undercounting in this age group, then IFR goes up further. This should be a cause for concern. 3/6
Some news reports on COVID-19 death undercounting in India... 1/
West Bengal: 50% of deaths (late March to April) were omitted on account of comorbidities before a "reconciliation" in early May. (Modelling suggests undercounting continued after this.) hindustantimes.com/india-news/cov…
Delhi. Many deaths went "missing" between mid-April and mid-May. Some were added back over the next month. (Modelling suggests at least 3/4 of deaths were missing at the worst point, and the reconciliation was probably incomplete.) thewire.in/government/del…
Tamil Nadu. Only about half of the COVID-19 deaths recorded by the Greater Chennai Corporation had made it to the state register by early June (scroll.in/latest/964371/…). More have since been going missing from the official count. (timesofindia.indiatimes.com/city/chennai/2…).
What is happening with COVID-19 in badly hit Delhi? Some observations from the data and modelling. Thread. 1/5
1) Cases are not slowing down. Doubling time over the past month has hovered around 14 days (its current value) - much worse than the national average.
2) Test positivity has been rising to dizzying heights (about 30%), which suggests strongly that "case detection" - the proportion of cases detected in testing - is falling. Simply put: testing is not keeping up with infection, but this could be changing now. 2/5
3) As least some (most?) of Delhi's missing deaths have been added back in. E.g., if we assume constant case detection, then we're almost there. With case detection down about 40%, there are still over 1000 deaths missing. (Second scenario is still conservative.) 3/5