Covid Seasonality

plausibly zero

Both lines have two points outside of the error bars

We have almost four years of data on covid in Canada, which is enough to start thinking about seasonality.

People who deal with data for a living, and who have analyzed thousands of data sets covering hundreds of different types of data, from text records to one, two, three, and higher dimensional numerical data, know that making a call with regard to cyclicity is not something that can be done lightly or casually. False cyclical signals come up a lot in low quality data analysis, from financial markets to influenza pandemics.

The astronomer Fred Hoyle once argued that the specific cycles of pandemic influenza relative to the Earth's motion through space pointed to an extra-terrestrial origin for the flu. Naively, the data looked not entirely wrong, but people with experience in cyclical analysis dug into them and showed the result was spurious.

As the absence of any actual public health leadership leaves covid seasonality an open question, it ends up on my plate to do this kind of analysis. I recognize that this is pointless: it brings no benefit to me, and my audience of just shy of 100 is unlikely to make up a large enough voting block to swing any elections toward parties that will make public health a priority. But I do it because I can.

In the Discworld novel Reaper Man, British author Terry Pratchett describes the bargain that the smiths of the kingdom of Lancre make: they can shoe anything that's brought to them, but there's a price… they have to shoe anything that's brought to them. It has something to do with iron. This is not a bargain I'm willing to make: a long time ago I decided there are some problems I am not going to work on, even when solutions came easily to mind. But I'm close enough to it to appreciate the reality, and in this case I'm kind of stuck with my own commitment to do what I can to tease out knowledge from the little data we have available.

So: what about covid seasonality?

The graph above shows the data we have: covid hospitalizations in Canada as a function of the month of the year, starting in January (month 1) and going to December (month 12). The error bars are large because the data is noisy: successive waves have been driven primarily by new mutants, not anything to do with human behaviour or seasonal cycles. We know this: the alpha and delta waves, and the many omicron waves since those innocent days, have all been variant-driven. So what we're faced with here is the question: do those waves tend to come seasonally?

To first order, the data are consistent with a steady level of about 3000 Canadians in hospital with covid at all times. This is shown by the flat line in the graph.

I've also fit a sinusoidal function to the data, and it's a slightly--but only slightly--better fit. The root mean squared or RMS error is the square root of the average of the squared differences between data values and fit values, and is a standard way to assess goodness of fit. For the sinusoidal fit it's 626, and for the steady level it's 879.

But the the standard deviation, which is a measure how variable the data are from year-to-year within each month, shows that this is a completely meaningless difference: as the error bars in the graph suggest, the standard deviation is just under 1200, which means the difference in quality between a seasonal fit and a flat fit is about 20% of the uncertainty in the monthly average due to year-over-year variation. That's the definition of "in the noise": there is no meaningful improvement in the fit on the assumption that covid is seasonal. None.

There is also a tendency for there to be a spring wave, if four years of data are adequate to talk meaningfully about seasonality. You can see that in the way the data jump above the "seasonal" fit in April and May (months 4 and 5 on the graph). But is that tendency an accidental occurrence? If someone wants to dismiss it as accidental, why not dismiss the comparably high numbers in January and December as accidental?

If we look at hospitalizations through time over the first four years of the ongoing pandemic, the peaks seen between "01" and "07" in 2020, 2021, and 2022 in the overall graph of Canadian covid hospitalizations are all in the April-May time range. Is that seasonality? Or accident?

On the other hand, we know the omicron peak was anomalous, so should we take it out of the data? Here's what happens if we do, removing December 2021 and January and February 2022 from the averages:

These data are also entirely consistent with a flat line, with about 2700 Canadians in hospital with covid at all times.

It's always a bit sketchy when we start culling data. There has to be a principled way of doing it, and "Well, the omicron wave was anomalous, amiright?" is not a particularly principled argument. Who's to say we won't get a comparable wave every other January, as immunity wanes and vaccine uptake falls and the virus evolves? We know all those things happen in a big way, and maybe something in the late autumn air in Canada--like being damp and cold, say--is enough of a trigger to set things off.

We do not have the data to rule that out. Nor can we rule out a wide range of alternative scenarios, at least not with anything like the degree of certainty we'd like. Only in the past year has covid curve settled down into something that looks vaguely smooth and well-behaved, and we have no idea how long that is going to last. We are currently in the eleventh wave in four years. In terms of area it's huge, and the peak is not yet in sight, thanks to the fantastically low data quality that is all the government of Canada can manage with all the tax money it gets from us.

To get a sense of what actually seasonal diseases look like, I've plotted some data for a variety of other diseases, in each case with the data normalized to 100%, so this graph cannot be used to compare the absolute incidence of the various diseases, only their pattern of infection throughout the year. I've included the flu on the covid graph at the top of this post to show how small it is relative to covid, at least in terms of hospitalizations. As well as being much less severe, it's also far more seasonal, as are RSV and colds caused by endemic coronaviruses. Each of these drops to practically nothing in the summer, after a pronounced winter peak.

Again: in absolute terms these have a fraction of the effect on health that covid does. As shown in the the graph at the top, flu hospitalizes people at a rate of less than 10% of the covid rate.

And as near as I can tell, these data are for actual flu, based on testing for the influenza virus, not "flu and pneumonia" , which is a meaningless category that only a physician could have come up with. It is meaningless because it mixes a cause (flu) with an effect (pneumonia) that has many other causes. Pneumonia is a symptom, not a disease. It's like reporting deaths by "car accident and blunt force trauma": that number would tell you almost nothing about how many people are killed in car accidents--which are just one of many causes of blunt force trauma--unless you know all the others. It's data theatre, creating the illusion of knowledge where there is none.

Unfortunately, for death rates, only "flu and pneumonia" is reported, meaning we have no knowledge of how many people are killed by influenza beyond an upper and lower bound, where the lower bound is zero. So we have no death rate for flu that can be compared to the death rate for covid.

As can also be seen in the graph of other diseases, some of them are not particularly seasonal. Unlike flu, et al, cold symptoms caused by parainfluenza and rhinoviruses are more common year-round, with parainfluenzas peaking late spring and late fall, and rhinoviruses in early fall, perhaps driven by school returning. Or perhaps not.

With improved ventilation--bringing schools, hospitals, office buildings, public transit, and other public places like theaters and seniors centres, up to ASHRAE 241 standards--we could cut the rates of these diseases down by orders of magnitude, which even ignoring the dent it would put in covid numbers would plausibly wipe out the eighth largest killer in the nation: influenza.

Put it this way: if you could wipe out deaths from suicides, would you? If we knew how? If we could just install a simple device in every public building that was nothing but an upgrade to an existing system and had no effect except ending deaths from suicides, would you?

The data say that the vast majority of Canadians would not, because covid kills at least ten times the number of people suicide does every year, and practically no one can be bothered to agitate for the clean air technology that would save them.

And I want to be very clear about this: it isn't that nothing is being done because the people dying are poor, or black, or indigenous, or gay. They're just people, like you, and covid is just as much a threat to you--today, if you're older; tomorrow, if you're younger--as it is to everyone else.

So anyone who says, "THEY would care if only..." is a blind to the data, and cares about nothing but using this disaster to further their own personal political agenda. But this isn't about politics. This is not "late capitalism's" fault, however much else plausibly is. People behaved the same way in the 1919 influenza, in the cholera outbreaks of the 1800s, and back in history through time and circumstance as far as we can see, all the way to the Black Death: after the first wave, people got on with life, not really minding if they themselves died of it, much less anyone else, regardless of their race, creed, class, sexual orientation, or number of legs.

The average person simply does not care.

Which is why it's important that we do.

And it's important that what we care about are the facts, as well as we can ascertain them.

If we don't, Canadian lifespan will continue to fall. The life expectancy of a 20-year-old Canadian in 2022 was about six months less than it was in 2019, which was the high point. It fell in 2020, in 2021, and in 2022, and the rate of decline has not yet slowed down: the drop from 2021 to 2022 is the biggest one yet.

So my conclusion in all this is that we do not have strong evidence that covid is a seasonal disease, and whatever seasonality there is, it appears to be weak at best. Treating covid as a seasonal disease in Canada at this time is unwarranted by the data, and talking about it in terms of "respiratory disease season" as if it was the flu or most colds is a lie.

We should not be taking public health guidance from liars and incompetents.

Controlling the spread of respiratory viruses is an engineering problem. ASHRAE 241 is the engineering standard that we need to implement to hammer respiratory viruses flat.

We will only do that if there is the political will to do so. I've written my MP, the PM, the Premier, health ministers, and so on... and those calls have fallen on deaf ears. Until enough ordinary people say loudly and clearly that they actually care about tens of thousands of Canadians dying every year from preventable aerosol-borne diseases, nothing will happen. No politician of any party cares about anything but their own power. No public health bureaucrat cares about anything but getting out of the office promptly at 3 PM on any given Friday.

It is only by making our political leaders uncomfortable and uneasy that will we see action on implementing ASHRAE 241: integrating it into building codes, and requiring conformance to it as an industrial and occupational safety standard.

Political leaders need to know that if they don't take action on implementing ASHRAE 241 that they will face a credible threat of being voted out.

Until then, we'll be stuck in a cycle of lies, complacency, and preventable deaths. Covid is currently responsible for between five and ten percent of all deaths in Canada. More than Alzheimer's, more than diabetes, more than accidents. More than everything except heart disease and cancer.

Unlike so many other leading causes of death, covid is one that we know how to address today. It doesn't take new vaccines. It doesn't take medical breakthroughs.

It takes cleaning the air using technology that has been available for decades.

So why aren't we doing it?