Thread by @MoriartyLab

Estimating actual and reported COVID-19 deaths in Canada in the coming months

🧵

It's Saturday and I have a little more time than during the week, so here's an explainer of how I'm estimating future COVID-19 deaths.

There are MANY yelling in my mentions that this is impossible.

Friends, it's due to exponential growth, which humans aren't good at estimating.

To estimate future deaths, you need a couple of key pieces of information.

1. The actual number of infections
2. The death rate for those infections

Let's start with the actual number of infections.

Most deaths reported until yesterday resulted from infections up to Dec 17, when we still had a roughly reasonable handle on how many infections were actually happening (always more than reported BTW).

Here are estimated actual infections in Canada from IHME and ICL models to Dec 13, courtesy of @OurWorldInData plus reported infections.

@OurWorldInData We don't have ICL estimates after Dec 13, but here's a plot of IHME-estimated infections in Canada to date.

You'll see that there's reasonable certainty about those estimates before Dec 18, but then it gets messy, and upper 95% confidence interval reaches 550K new infections/day

Remember, the week of Dec 18 was the week before Christmas.

People were frantically trying to find rapid tests.

Out-of-home mobility was high, especially in grocery stories, pharmacy and retail, and it more than offset drops in work-related mobility and use of public transit.

That week, many people who should know better were still telling people they could mix with others if they used rapid tests, masking etc.

To be fair, it's not clear people would have limited indoor gatherings to those in their household even if they had been clearly warned.

Around that week, test positivity started rising fast, and there were many reports of people being unable to get PCR testing or rapid tests.

Estimates of actual cases in that week are shaky at best because so many switched to rapid testing, if they had access.

Then Christmas week happened, and in Canada, we have really problematic and delayed reporting from Dec 23 to the end of the first work week in January/middle of the second January work week.

So, things went dark at a really critical time.

Then, of course, after the holidays many Canadian regions changed their testing policy to prioritize only testing of symptomatic people at greatest risk of severe outcomes.

So actual case estimates are shaky and are likely underestimates.

DYK Canada is only high income country in the world that limits testing to symptomatic, high priority people?

Other countries that do this:

Aruba, Cambodia, DRC, Fiji, Guyana, Indonesia, Kiribati, Kyrgyzstan, Mexico, Palestine, Rwanda, S Sudan, Suriname, Syria, Turkey, Yemen

All of this is to say that estimates of cases in Canada are quite likely under-estimates. We're charting new ground for a high income nation.

Also, the surveillance that @GovCanHealth normally does is currently not happening because of overloaded testing capacity.

@GovCanHealth Alright. Let's go back to that IHME estimate of cases showing that were possibly as high as 550K cases/day on January 7.

How correct is it from Dec 18 onward?

Since cases are hugely under-reported, and we have provinces down-revising hospitalization numbers it's hard to know.

@GovCanHealth Yesterday's @GovCanHealth modelling presentation suggested the IHME estimates may be too high. Since I can't access the daily numbers for this modelling, I've taken the average of the IHME and ICL estimated actual case numbers, which gives something closer to the PHAC estimates.

@GovCanHealth After Dec 13 (last data available for ICL), I've assumed that ICL estimated infections were 2X reported infections and taken the average of this estimate and the IHME estimate to get estimated infection numbers that look more like PHAC estimates.

@GovCanHealth OK, that's how I estimated cases. We can test these estimates using an imperfect ballparking method based on reported deaths, which I'll get to shortly.

Next, estimating expected infection fatality rates and deaths.

This is done separately from trying to guess infection numbers.

First, I calculated the age-specific infection fatality rates we'd expect in unvaccinated people if Omicron is 50% "milder" than Delta.

Then, I calculated the expected rate of breakthrough infection death in vaccinated people, using the odds ratio published in the @GovCanHealth daily epidemiology update PLUS observed rates of vaccine protection against any infection reported daily by @COVIDSciOntario .

Vax protection against any infection is for people with min 2 doses, so it includes boosters.

FYI, current vax protection vs death w/ Omicron is 76%, due to some evasion of vax immunity. That's still high, but if you infect millions of people breakthrough deaths add up fast.

Then I estimate the maximum % of the fully vaxxed population that can currently be infected with Omicron from the @COVIDSciOntario protection vs any infection numbers.

Thanks to boosters, yesterday it was 59.1% in Ontario, a big improvement from the 85% reported on Dec 28.

@COVIDSciOntario Using the most recent full vaccination rates in each age group reported by PHAC for each province, PLUS the July 1, 2021 StatsCan population estimates for each age group in each province, I estimate how many total deaths we'd expect among unvaccinated people if 100% are infected.

@COVIDSciOntario Then I do the same for fully vaccinated people, assuming an upper ceiling of total people who can be infected based on the ON Sci Table estimates (yesterday, about 60%), and the current estimate of vax protection vs death (76%).

@COVIDSciOntario So, the expected total deaths among vaxxed people is the infection fatality rate for Omicron in each age group x 0.6 x 0.24.

Then I add up the expected total deaths for both vaxxed and unvaxxed people I get from these calculations.

@COVIDSciOntario Finally, I subtract the total deaths reported in each region since July 1, 2021 to control for deaths that have already happened.

Actually, for provinces outside Quebec I subtract a multiple of the reported deaths from July 1 to present to control for under-reporting.

@COVIDSciOntario Finally, I divide prior death-adjusted number by Jul 1 pop estimates to get a global estimated case fatality rate.

You can find estimated total future deaths here. They change daily due to changes in vaccination, vaccine protection, reported deaths.

docs.google.com/spreadsheets/d…

@COVIDSciOntario So, coming back to the graph I posted at the very top of this thread.

Here it is again.

The way I get the Expected Actual new daily deaths is by multiplying the global case fatality rate described above by the estimated infections on the day 4 weeks before the death.

@COVIDSciOntario The expected actual deaths are shown as a dashed green line. The light green lines projecting above and below that line are the 95% confidence intervals of that estimate, based on the 95% confidence intervals for estimated actual infection numbers 4 weeks previously.

@COVIDSciOntario What are the Expected Reported new daily deaths?

This is how many COVID-19 deaths will actually be detected and reported.

@COVIDSciOntario How do I estimate how many deaths will be detected/reported?

This depends on knowing what proportion of excess mortality in each region is accounted for by reported COVID-19 deaths.

Canada being Canada, we still don't have good excess mortality estimates past November 2020.

@COVIDSciOntario Reported COVID-19 deaths account for all excess mortality in Quebec to that date (and still do).

Ontario reports 2.1X fewer COVID-19 deaths than expected from excess mortality.

Manitoba: likely 1.7-fold under-detection/reporting.

Saskatchewan: 7.3X

Alberta: 3.4X

BC: 4.9X

@COVIDSciOntario Overall, COVID-19 deaths outside Quebec are likely under-detected/under-reported by 2.5-fold, and Canadian COVID-19 deaths are likely under-detected/under-reported by about 60%.

BTW that scale of under-detection/reporting is common in many countries, including most OECD peers.

@COVIDSciOntario Who care?

Because if we want to test if underlying assumptions about infection numbers, fatality rates, future deaths are reasonable, we need to compare to reported numbers.

So, we have to correct for under-detection/under-reporting.

@COVIDSciOntario In the graph here, the EXPECTED REPORTED new daily deaths are shown as a black dashed line, with gray confidence interval bars.

And in pink, you can see the 7 day moving average for new daily REPORTED COVID-19 deaths in Canada.

Solid match between pink and black curves so far.

@COVIDSciOntario That match is good up until yesterday, but remember, from this point on the deaths we'll be seeing are the ones from infections contracted after Dec 18, when new daily infection estimates are likely pretty messed up.

@COVIDSciOntario If I were to bet, I'd guess that we'll start seeing the pink line rise above the black dashed line, since there's a good chance infections are underestimated.

I don't think it will go down, unless provinces next start down-revising reported deaths thinking they're incidental.

@COVIDSciOntario We'll be able to see whether something like this is happening by comparing Ontario and Quebec expected deaths over time. IHME has infection estimates for these provinces, but there are not ICL provincial estimates, so I'll have to think about how to do this.

@COVIDSciOntario Finally, if you're reading this thread to the end and are a decision maker/influencer in your province who may be arguing that some deaths with COVID are not COVID deaths.

1. Remember WHO, PHAC guidance--it's a C19 death unless there's a clearly unrelated cause like trauma

@COVIDSciOntario 2. When excess mortality numbers finally come out and you're currently splitting hairs about C19 death reporting, there is a VERY good chance you'll find you have a lot of excess mortality that can't be explained by any specific thing. And this won't happen in Quebec.

@COVIDSciOntario 3. Finally, to those who think their clinical judgement about cause of death makes them better able to determine if someone died of or with COVID-19, remember that only by accurately reporting data can we determine this retroactively (god forbid Canada would do this in real time)

@COVIDSciOntario 4. COVID doesn't have to present as respiratory failure to kill someone or contribute to death. When someone w/ health problems contracts influenza and dies, do you say the influenza killed them or was it other underlying health issues that contributed to influenza vulnerability?

@COVIDSciOntario 5. Parsing all of this out requires statistical analysis of large populations. Let others learn more quickly about what is happening in different populations, what the risks and outcomes are. Without complete, accurate reporting Canada has to rely on research from other countries

@COVIDSciOntario If you're someone who focused on Omicron mildness and didn't think through implications of loss of vax protection for deaths in Canada, please don't finesse the deaths that are coming by saying some are incidental.

At minimum, the people who die(d) deserve to be counted.

Finally, want a full discussion of reported COVID-19 deaths and excess mortality in Canada?

Here you go: rsc-src.ca/en/covid-19-po…