Thread by @MoriartyLab

A thread about COVID-19 and COVID-19 reporting in Canada, as well as about validation of our latest COVID-19 Forecast model.

It's partly motivated by the latest infection estimates for Canada.

As you can see, estimated daily infections in Canada are currently at or nearing the highest Omicron peaks to date, and there's not a fresh vaccine in sight @GovCanHealth.

We URGENTLY need a decision about vaccine approval and shots in arms yesterday, not in October, which looks like it will be far too late, unless infections keep going even higher.

The Forecast comes out tomorrow.

I'm just posting a bit of a rant/technical discussion in advance, so that @BattlingBeaver doesn't need to deal with it when she tweets the regular Forecast thread for me tomorrow. Now she can just link to it in the main Forecast thread.

This thread is mainly an update on where we are with predicting excess mortality from waste water, our model and Forecast scores.

We're near peak predicted excess mortality for Canada since 2020 right now. We can't get new vaccines soon enough. I can't keep saying this often or loudly enough. @GovCanHealth we BADLY need the new vaccines.

About 1,000 people per week are dying from COVID-19 this week in Canada. We cannot get vaccines soon enough. Please. And provinces, we cannot delay vaccine rollout until October. This is so urgent.

In the graph there are a lot of different lines.

All are showing % excess mortality (plotted on the left axis) and % excess mortality expected from reported COVID-19 deaths (plotted on the right axis).

The scale and unit (% excess mortality) is the same for all of them, so you can compare lines directly.

This is a bit silly. Now that I've had to pay for Twitter/X, I wanted to edit the post I just made, and can't figure out how to do it. I could do this last week. Maybe I'm just not seeing it easily.

What I meant to add was that I really need to switch the plotting so all values are plotted on the left y-axis, since there's no need for a right axis. But, some things take a long time to get done on my watch.

Oh hey! I just saw that I can add bold and italics too. Yay.

This graph shows data for the Canadian provinces that report all-cause mortality fastest (Newfoundland and Labrador, Quebec, Alberta, British Columbia).

We use these provinces because excess mortality estimates from Statistics Canada are more complete, which means they're better for testing predictions.

Even for these provinces the most recent 6-12 months of estimates available from @StatCan_eng are still incomplete and will increase.

Unfortunately, due to a pause in the StatsCan weekly excess mortality program, estimates are not available past September 2, 2023, even for fast-reporting provinces.

We will transition to doing our own modelling if the program remains on pause for much longer, but really don't want to, because StatsCan estimates have been outstanding because they adjust for the number of expected deaths that have not yet been reported.

This means other widely available excess mortality estimates from @TheEconomist and @OurWorldInData catch up to StatsCan a year or two later (largely because of very slow reporting by Canada's largest province, Ontario).

StatsCan has a lot of data available to them related to population growth, aging etc.

It's unlikely we or anyone else could do excess mortality estimates for Canada and Canadian provinces as well as they do, so we're loathe to do this ourselves, because it likely just won't be as good.

And it will take us months to set up and test a model.

In the short term, we're estimating excess mortality for Canada from weekly reported COVID-19 deaths in Quebec, adjusted by the average excess death: reported COVID-19 death ratio for the same week in the preceding two years.

However, because death detection/reporting continues to decline, we really need more recent data to correct reported deaths properly.

We can do this when annual mortality data come out in December of each year, but we won't get 2023 numbers until this December.

And we also use data from the CVS-D, CIHI and the PHAC case dataset, but reporting to/by all of these sources is usually delayed by at least 6 months.

Weekly excess mortality estimates from StatsCan were really important to give a more timely source of information.

@BattlingBeaver Alright....that was quite an excursion from the main topic, which is explaining the graph showing prediction of excess mortality from waste water, our model and the Forecast.

The Forecast scores are scaled to correspond to % excess mortality.

A score of 0-5 (moderate) corresponds to 0-5% excess mortality.

For Canada, 5% excess mortality (Forecast score of 5) corresponds to ~276 excess (untimely, "from" not "with") deaths from COVID-19 per week.

A Forecast score of 10 is ~552 "from" COVID deaths.

A Forecast score of 15 is  ~828 untimely COVID deaths/week.

A Forecast score of 20 is 1,1104 COVID-19 deaths/week.

A Forecast score of 25 is 1,380 deaths/week.

The scale of our Forecast goes up to 20, because anything above 15 is obviously a severe situation, and we don't want to have a category for extra severe (what's the word you would use?).

Severe means you REALLY need to take precautions, no matter if it's severe or extra severe.

We never really get to a LOW situation in Canada where excess mortality is zero, so we've grouped scores 5 and lower into a single group called MODERATE.

This seems to be as good as we get, unfortunately.

The moderate range is 1-5 times higher than what we'd expect for annual fluctuations in deaths from influenza, which are lower and higher in different years.

If you're interested in numbers of untimely COVID-19 deaths for your province, you can just go to the deaths pages of our biweekly report.



Or you can multiply the numbers for Canada above by the size of the population in your province.

It's not quite as simple as this, though, since the infection fatality rates are different for different provinces, due to population age, underlying health, vaccination status, healthcare system effectiveness etc.

For example, the infection fatality rate for Newfoundland and Labrador is the highest in the country (49% higher than the Canadian average).

The Rock has an older population with higher rates of underlying health issues and lower hospital heart attack survival rates than the Canadian average (hospital heart attack survival rates is an indicator of healthcare system performance).

Some provinces with particularly high infection fatality rates may also hospitalize fewer people with COVID-19 than others, resulting in lower survival rates on a population level.

Others may by more proactive about limiting spread of infection in high risk settings (healthcare settings, longterm care, retirement homes).

For example, some jurisdictions may require masking in these settings earlier in surges than others, if they require this at all.

The lowest infection fatality rate in Canada is in Alberta, which is Canada's youngest province. It's about 11% lower than the national average.covid19resources.ca/covid-hazard-i…

Canada's highest infection fatality rates are typically in the summer/early fall, before new vaccines become available, when people may not realize there's a lot of COVID-19 around, when hospitals and clinics may not be doing as many COVID-19 tests because it's not traditional respiratory infection season and because tests are sometimes done in fall/winter to get a differential diagnosis (flu and COVID treatments differ).

We also have big summer/early fall surges each year in Canada, like in the United States, coinciding with peak air conditioning season when people may move indoors again.

If you're interested, in the next post I'll share the estimated Omicron population infection fatality rates (pIFRs) for each province up to September 2, 2023.

pIFRs for particularly slow-reporting provinces (Ontario, Nova Scotia, Manitoba) are likely under-estimated.

You can multiply these pIFRs by the estimated number of infections per week in your province to figure out how many deaths are expected.

The pIFR fluctuates a bit over time. Right now it's near its peak for most provinces because we're waiting for fresh vaccines, and the old vaccines aren't as good a match for the currently circulating variants.

So multiplying by the global pIFR for your province doesn't always perfectly predict deaths at all times of year, but usually the pIFR fluctates by not much more than 1.6-fold, so you multiply or divide your estimates by 1.6 to get a range.

You can do the same for hospitalizations (or go look at the hospitalizations page in our weekly report!).

The UK reports just over 5 hospitalizations for every death.

France reports nearly 10 hospitalizations for every death.

The number of people needing hospitalization is probably closer to the value for France (a 10% in-hospital mortality rate, which is still more than 2X higher than normal in-hospital mortality rate for other causes of hospitalization).

But, since Canada and the UK have many fewer hospital beds per capita than France, we many actually only admit half the people who need hospitalization--so 5 hospitalizations for every death.

A bit more on this later.

Estimated pIFRs by province, highest to lowest:

--CAN 0.0903% (0.0903% of people infected die an untimely death from COVID-19).
--NL 0.1343%
--NB 0.1306%
--SK 0.1241%
--MB 0.1140%
--PEI 0.1124%
--BC 0.1025%
--NS 0.0872%
--QC 0.0864%
--YT, NT, NWT 0.0855%
--ON 0.0832%
--AB 0.0802%

There were/are about 1.3 million infections in Canada this week.

Multiply that by 0.0903% to estimate about how many COVID-19 deaths will be expected in the coming weeks from infections this week.

It's about 1,200.

@BattlingBeaver OK. Back to the explanation of the graph.

Sigh. I'm terrible for digressions.

In the graph, the bright pink line is % excess mortality each week (Statistics Canada, corrected to remove non-COVID-19 causes of excess death such as toxic drugs).

The gold line is % excess mortality predicted from our model.

The teal line is % excess mortality predicted from waste water alone.

The black dashed line is % excess mortality predicted by the Forecast score.

The solid purple line way down at the bottom is the % excess mortality calculated from weekly COVID-19 deaths reported on provincial sites.

The dashed light purple line is the % excess mortality calculated from weekly COVID-19 deaths reported in all sources (provinces, the Canadian Vital Statistics Death database: CVS-D), the PHAC individual case dataset released quarterly by StatsCan and in-hospital COVID-19 deaths reported annually by @CIHI_ICS.

By comparing the solid purple line (deaths reported on provincial sites) to others, you can see just how few COVID-19 deaths in Canada are reported in a readily publicly accessible fashion.

You can compile data from a bunch of different sources, as we do for the data shown in the dashed light purple line.

I suspect most people living in Canada don't have the technical skills or grim patience required to assemble this puzzle on a regular basis.

And don't want to and shouldn't have to.

After all, if it was important, wouldn't provincial health ministries make it easy to find and interpret the data for something so important as the health of provincial residents?

Or at least report the data to PHAC, since COVID-19 is a reportable infection and PHAC is the federal body to whom COVID-19 health data should be reported by provinces?

These are reasonable assumptions for people to make.

And cynical exploitation of these reasonable assumptions is pretty anti-democratic.

Anyway....not that health data matter.

If they did, provinces wouldn't spend so much time refusing to share it with the public or with federal bodies, and fighting to avoid sharing it.

We need more journalists like @robyndoolittle, but it is a painful beat requiring so much forensic data sleuthing and grim determination.

And how many journalists are left in Canada with the time required for this investigative work.

The painful inexorable decline of public interest journalism and healthcare and health data reporting in Canada go hand in hand.

@BattlingBeaver These segues are probably impinging on your very last nerve.

The graph also shows the highest and lowest % excess mortality values since February 2020 to provide a sense of scale.

These are the horizontal red dashed line at the top of the graph (max) and the horizontal blue dashed line at the bottom of the graph (min).

The LOWEST % excess mortality (~2%) was observed in the late spring/early summer of 2021, just as the most people at greatest risk had received at least one vaccine dose (many/most with 2 doses).

Vaccination continued to increase dramatically after this point, but so did infections, as people started dropping other protections.

And of course later summer/early fall of 2021 was the period of the Delta variant, which had substantially higher infection fatality and hospitalization rates, especially for people who were unvaccinated or who were infected immediately after vaccination, before vaccine immunity had developed.

The HIGHEST % excess mortality in the pandemic to date was late November/early December of 2022 (although we don't have data past September 2, 2023, and it's likely excess mortality has gone as high or possibly higher than this maximum since then).

Vaccination rates really started plummeting in the summer/fall of 2022, because people were getting infected before they got their new doses.

The two main takeaways from this graph are the following:

1) Provinces are only reporting a small fraction of actual COVID-19 deaths on the most accessible and widely accessed public reporting pages.

Overall, fewer than 20% of COVID-19 deaths are being reported in these public places in all provinces (a bit higher in Quebec in fall/winter season, but not much).

2) The model and waste water and the Forecast score are good at predicting excess mortality and COVID-19 deaths.

Over time, as more reporting comes in via various sources, the purple dashed line and pink line typically catch up to the prediction (although not completely for the purple dashed line, because some deaths are likely missed since testing has declined dramatically).

There's always more wobble in more recent months, before slow-reported data come in that help us re-calibrate assumptions about under-reporting rates.

But it's generally fairly stable.

So....

When we say the score is 20 and the Forecast word score is SEVERE, this means about 1,100 people in Canada are dying untimely (excess) deaths each week.

This is about 20X more deaths than we'd expect from influenza.

At a score of 20, COVID-19 is likely the number 1 or 2 cause of death in the country that week.

That seems pretty severe to us, although word choice can be value-laden, and if it's severe for months on end, which it often is from now to the end of December, then the value of that word score may be diminished.

It's a bit like discussions of climate change, as we break more and more historic records.

How do we talk about this without people tuning out, but also being accurate in our description?

It's a major problem, and one that is cynically politicized by those with incentives not to act or respond when collective action is needed.

Provinces have really got themselves into a bind by the impressive scale of under-testing and under-reporting.

As you can see from p. 26 of our biweekly report, ~1,500 people/day are likely being hospitalized with COVID-19 this week in Canada (10,500 this week).

Canada is likely spending $37.4M/day on COVID-19 hospitalizations this week (based on CIHI cost estimates).

Canadian hospitals are likely 18% over-capacity this week due to hospitalizations of people with COVID-19 (whether these are detected by testing or not).

These estimates are based on the most recent hospitalization rates per death reported by France, which reports the greatest proportion of its COVID-19 hospitalizations of G7 countries.

And no, they're unlikely to be incidental "with" COVID-19 hospitalizations.

The mortality rate per hospitalization in France is still two times higher than the average in-hospital mortality rate for all diseases/conditions.

The UK reports about half the hospitalizations of France.

Like Canada, the UK still has stubbornly high excess mortality rates (if you look at data from the Economist or OWID, which don't incorporate 2020 into their baseline).

Excess mortality in 2023 was as high (or higher) than in 2022 in the UK (Economist estimates), and it looks like this was true in Canada too, based on Statistics Canada data for fast-reporting provinces).

Our estimates of infection hospitalization rates are closer to France than the UK, because the UK likely under-reports hospitalizations by quite a bit.

But.....it's possible that in the UK this is not exclusively an under-reporting issue, but may also reflect lower rates of hospitalization for COVID-19 (and other conditions) than in France, which has many more hospital beds per capita than the UK (and Canada).

Canada and the UK have the lowest number of hospital beds per capita among medium-large high income democratic peer countries.

We're similar to the UK in many respects.

We're slightly older than the UK, have slightly fewer hospital beds per capita and have lower rates of fresh vaccination since 2021.

Our median weekly pIFR from December 2021 to March 2023 estimated from excess mortality is similar to the UK SARS-CoV-2 pIFR for the same period (1% higher than median for UK), as is estimated weekly prevalence of SARS-CoV-2 infections.

In both the UK and Canada, excess mortality rates remain stubbornly high--much higher than France and higher than the United States, which also reports higher hospitalization rates.

The question is, why are the pIFRs for the UK and Canada still so high?

Does this reflect less access to hospitalization for people with COVID-19 in these countries, and perhaps also for people with other conditions, because we have so little surge capacity and because the ongoing burden of COVID-19 on the acute care system is so high?

Do France and the United States report all-cause mortality more slowly than the United Kingdom (yes...they do. The UK is one of the fastest-reporting countries for all cause mortality).

Provinces are in a bind because of under-reporting of COVID-19 for a bunch of reasons.

First, uptake of fresh vaccines is abysmal in Canada--lower even than uptake of flu shots, and now lower than in most peer countries, including the United States.

This is likely partly due to the policy of not recommending shots if you've been infected in the last 6 months, and the absence of spring vaccination campaigns, as well as slower approval and rollout of new fall vaccines in the late summer/early fall months when they're most needed.

We need to be vaccinating as fast as the United States, which started last week.

If vaccines don't become available to people in Canada until October we're about 6 weeks behind, during a period when infections are at a peak and when we need vaccines most.

We need new vaccines by mid-summer, frankly.

However, a lot of people aren't getting fresh vaccines anymore because they think COVID-19 is a lot better right now and that it's not as urgent to get vaccinated.

This clearly includes many people at high risk, as is evident from age-specific vaccination rates.

A key barrier to vaccine uptake is complacency, or lack of concern about the disease for which vaccines are available.

Truly shameful levels of under-detection/under-reporting of COVID-19 hospitalizations and deaths by all provinces directly contributes to complacency and poor vaccine uptake, even among those who need them most.

The other reason provinces are in a bind is because COVID-19 is still a relentless problem for hospital capacity and spending, including in the summer when staffing levels are lowest.

However, if you don't report numbers fully in publicly accessible places, or if you don't even report most of your hospitalizations and deaths to PHAC, then it gets hard to ask for more money to fund hospitals even though you badly need it.

If you bring alternative numbers to the table that show just how high the burden of COVID-19 is, and if those numbers differ substantially from the numbers reported in the most public places, then policy and decision makers at the federal level will rightly question which numbers are true, and if internally-presented numbers are exaggerated to justify requesting more money.

Finally, there's not a province in this country that isn't struggling with ER closures and extreme hospital bed and family physician shortages, as @alandrummond2 and other clinicians and hospital leaders can (and sometimes do) attest.

If provinces actually reported COVID-19 data more fully, it would make it easier to explain to the public why this is such an ongoing issue.

And yes, it's also an issue of shortages of healthcare workers, particularly nurses and family doctors.

There can be two simultaneous problems, or in the case of Canadian healthcare, many simultaneous problems.

But given what the healthcare workforce has gone though since 2020, and is still going through, wouldn't you be planning your exit or moving to agency employment or private clinics or reducing family practice hours or taking early retirement?

We urgently need to retain the healthcare workforce, and that means not permanently operating hospitals and other healthcare services at or above crisis capacity.

People just cannot sustain working at this level for years on end.

And Canada and the UK are the worst of peer countries in this situation, because we had razor-thin surge capacity even before COVID-19.

On average, since the beginning of Omicron, people needing hospitalization for COVID-19 account for 14% of hospital bed capacity (7% if you admit only half of people needing hospitalization).

Even 7% overcapacity on an ongoing basis is extraordinarily challenging.

Think of just-in-time manufacturing and shipping. Think how the blockage of one trade route by the Evergreen (was that the name?) affected markets and supply of good worldwide.

Think of how a volcano in Iceland affected air travel everywhere and for how long.

These are all systems, like our healthcare system, that routinely operate as efficiently and with as little room to spare as possible.

A constant annual 7% increase in hospital beds required for COVID-19, in a very low surge capacity environment with a serious healthcare workforce labour shortage, can have profound upstream and downstream effects on healthcare and health.

And that, my friends, is the end of this weird hybrid technical/ranting post.

Have to go back to grant writing.

Although I post about grim things a lot, remember that hope and faith in humanity drive change.

I care about you and your life, just as you care about others and their lives.

We're human beings. We get overwhelmed, but we care.

Sometimes when we feel overwhelmed we become desensitized and feel powerless, and we detach from issues that we do in fact care about. For ourselves, for those we love, but also for other people in the abstract.

But we do matter, and the collective effect of many individuals trying to do one thing to help and make a difference can be powerful.

Keep loving each other and looking out for each other and remembering that every single person is imperfect and inconsistent, but most of are fundamentally kind, decent people who care.

That's who you are and I am and who we are.

I'm tired of the shit-disturbers who want to drag everyone down into their nihilism and misery.

I believe in people, and in you.

Much love.