Drowning in Data

A study by Flacco et al looked at potentially vaccine-related serious adverse events (PVR-SAEs) in a single province in Italy. It looks like the authors forgot to include some deaths in their calculations, and I’ve also noticed a few issues with timings and cohort sizes.

In brief:

• Every vaccinated cohort is compared to the entire unvaccinated cohort. Their sizes and follow-up times do not match, which casts doubt on the validity of the Cox relative risk calculations in the paper.
• The risk of PVR-SAEs are very high in the ‘1 Dose’ and ‘2 Doses’ categories compared to unvaccinated people. The supplementary data table shows how high the risks are for these vaccinated groups.
• Deaths for the ‘Greater than one dose’ group reported in the data tables don’t match the deaths in the Kaplan-Meier plot. When those deaths are considered, this group has approximately the same risk of all-cause death as the unvaccinated cohort (which disagrees with the authors’ calculations).

An important step in comparing different treatment groups in any study of drug effectiveness is matching follow-up times and group sizes. In this case, the authors were using all of the health records from one province in Italy (page 2) and were conducting a retrospective cohort analysis. Unlike a randomised controlled trial (RCT) where two matched patient groups are recruited and followed up for a set time, the authors used all of the health records from a single province between 2 January 2021 and 31 July 2022. This places some limitations on the data available. The authors couldn’t keep on recruiting patients until a target cohort size was reached. They had to work with what they had from the whole population in the province.

One problem is that the sizes of the different groups are not well matched (see graph below).

Another limitation is that the follow-up times are different. Note that the authors use the term “follow-up time” to mean “time to event” – the time measured from the start of the study until an event happens that fits the PVR-SAE criteria. The follow-up times (times to event) varied across each cohort.

This graph is based on the information in Table 1 (page 5). Curiously, it shows that the majority of unvaccinated people suffered some kind of PVR-SAE near the end of the study at around day 561. The standard deviation of 111 days takes these follow-up times to beyond the end date of the study.

The biggest problem I see with the way the authors have compared different groups is that all of the unvaccinated cohort are compared to all of each vaccinated group. This is show in the flowchart from the paper (Figure 1, page 3) below.

This means that instead of considering just the unvaccinated from July 2021 until the end of the study for the ‘Greater Than or Equal to 3 Doses’ group, all of the unvaccinated people from before this vaccine group existed were also used to calculate hazard ratios.

One assumption of the Cox method used by the authors is that of proportional hazards. It appears to me that comparing the unvaccinated group to the ‘Greater than one dose’ group breaks that assumption. Hazards will change across the period of the study in that vaccine group. I think that comparing ‘1 Dose’, ‘2 Doses’ and ‘Greater Than or Equal to 3 Doses’ to only those unvaccinated people who were recorded during those three different time periods would have made a more useful study.

Another Cox assumption is that the length of the study is the same for each group considered. As discussed above, this assumption is broken for all except the ‘Greater than one dose’ group. More information on problems with how Cox survival analysis has been used in other studies on COVID-19 treatment is discussed by Piovani et al.

The authors state

…the likelihood of none of the individual PVR-SAEs was significantly higher among subjects who received at least one dose of vaccine…

page 10

and

…all subjects who received at least one dose, showed no increase in the incidence of any outcome…

page 11

yet the data tables show that for the ‘1 Dose’ and ‘2 Doses’ groups, this was not the case. The hazard ratios for each group are graphed below. These data are taken from Table 3 on page 9 of the paper.

In 10 out of 14 outcomes, the ‘1 Dose’ group had larger hazard ratios than the unvaccinated (up to 20x larger for myocarditis or pericarditis!). In 11 out of 14 outcomes, the ‘2 Doses’ group had larger hazard ratios than the unvaccinated.

The ‘Greater Than or Equal To 1 Dose’ and ‘Greater Than or Equal To 3 Doses’ groups always had lower risks of every PVR-SAE than the unvaccinated group. However, the authors may have forgotten to include some deaths in the ‘Greater Than or Equal To 1 Dose’ group.

The authors include a Kaplan-Meier (KM) plot (page 8) showing the survival estimates of time to death for this group and the unvaccinated group. The line on the KM plot labelled ‘One or more vaccine doses’ shows that 98.3% of this group survived to the end of the study. This is compared to 95.8% of the unvaccinated group.

This means that 4417 (1.8%) of the ‘Greater Than or Equal To 1 Dose’ group died during the study according to the KM plot. This disagrees with the results in Table 2 (page 6) of the paper which states that 3351 (1.29%) of this group died.

I read the data from the KM plot using WebPlotDigitizer to extract the times of deaths from the unvaccinated and ‘Greater Than or Equal To 1 Dose’ groups. I did a Cox Regression to calculate proportional hazards between the two groups using R (with the survival package). My results show that there was no significant difference in the risk of all-cause death between the two groups.

Risk of all-cause death in the ‘Greater Than or Equal To 1 Dose’ group is 1.07 (95%CI 0.99 – 1.15) compared to the unvaccinated group. The confidence intervals contain a hazard ratio of 1, therefore there is no difference between the two groups. My result with the extra deaths included is different to the authors’ result – they said that the hazard ratio for all-cause death in this group was a far lower 0.19 (95%CI 0.18 – 0.20).

Note that the total size of the cohort in my analysis (n=316329) differs from that in the paper (n=316314) by 15 people due to slight errors in the graph reading process when using WebPlotDigitiser. This makes little difference to the overall result. If the authors released the raw data (I’ve asked with no response from them yet) that would be very helpful in understanding this more.

The authors define “vaccinated” as “vaccinated”.

…the follow-up started the day of the first (or single) dose of vaccine for vaccinated subjects…

Flacco et al, page 2

There is no “two weeks to let the vaccine take effect” waiting time before someone is considered to be vaccinated. This is the first study I’ve seen that does this – every other study I’ve heard of uses two or three weeks after a vaccine dose before someone is considered vaccinated.

The authors finish their abstract with this statement:

Further research is warranted to evaluate the long-term safety of COVID-19 vaccines.

Flacco et al, page 1

Sadly, this is the main problem with all studies on COVID-19 vaccines. Long-term safety data should have been known before any vaccination programme began. The study reviewed here ignores a massive safety signal in the greatly increased hazard ratios for the ‘1 Dose’ group, and glosses over them by considering the whole of the vaccinated group instead.

It’s clear from this study that COVID-19 vaccination carries an increased risk of various types of circulatory and respiratory adverse events. It’s unhelpful that this study has inconsistencies in the data used and how it is reported. This appears to be a really promising study that would help to understand how frequently PVR-SAEs occur following vaccination. The missing deaths data and the problems with the authors’ analysis make this a missed opportunity to help us to understand COVID-19 vaccine safety.

Well, it’s been a wee while. I’ve had a busy year, but it looks like the various excess deaths problems I looked at a year ago haven’t gone away. Why should we bother looking at excess deaths anyway? Are these excess deaths a problem?

There’s a lot that’s been written about excess deaths and what they mean. One critical article on their use by Norman Fenton and Martin Neill states:

…one of the standard approaches for determining whether increased death counts are likely explained by COVID and/or lockdowns is to use the previous 5-year datasets of death numbers and the model ‘fitting’ approach described above…you need to be very wary if you see conclusions about the current week, month or year death numbers being ‘significantly above the 5-year average’. 

https://wherearethenumbers.substack.com/p/what-can-we-learn-from-very-few-data

However, there have been various reports in the media and from the Office for National Statistics (ONS) on excess deaths and their increase over the last two years. Opinions differ about their use, but my main argument is: if the various interventions intended to prevent death from COVID-19 have been effective, shouldn’t the all cause excess deaths in 2021 and 2022 show a decrease compared to 2020? If we can see a reduction in all cause deaths in the last two years, this implies that the various COVID-19 interventions were effective and saved lives, and that the vaccination programme hasn’t caused further deaths.

Of course, it’s all a bit more complex than that due to various other economic and political factors. That aside, what do the raw data say about the excess death situation in Scotland?

First off, some background on the data. The reason I’m sharing these more detailed excess deaths data is because Public Health Scotland (PHS) on their dashboard only shares excess deaths for two age ranges: under 65 and 65 and above. This doesn’t show the full range of detail available in the deaths data, and may miss important changes in specific age groups.

I’ve published the code I used to calculate excess death and the data on my Github page.

I used two datasets supplied by National Records of Scotland to calculate excess deaths:

1. “Weekly deaths by sex and age group, 2000 to 2019” which was used to calculate the baseline data using weekly mortality by age group from 2015 to 2019
2. “Weekly deaths by sex and age group in NHS health boards, 2020 – 2022” which contains weekly deaths for 2020 to 2022 divided by cause.

The age categories are in (mostly) 5 year groups which can be seen in the weekly deaths graph for 2015-2019 shown below.

I’ve put each age group in a subplot. The range of weekly deaths is quite different for each age group – check the scales on the vertical axes before making detailed comparisons.

The 2000-2019 deaths data have different age categories to the more recent deaths data. I’ve combined the 90-94 and 95+ age ranges from the 2000-2019 deaths data into a single 90+ age range. This makes the age ranges of both datasets compatible.

The ONS changed the method used to calculate excess deaths in 2021 and 2022. Typically, deaths data from the previous five years (from 2015 to 2019) would be used to calculate a baseline for the next year. However, due to the larger number of deaths in 2020, their calculations for 2022 use deaths data from 2016 to 2019 and 2021 instead. Out of curiosity, I’ve calculated excess deaths for 2022 using both age ranges (just 2015-2019 and 2016-2019 plus 2020). Both datasets are on Github and the baseline results are shown below.

In most cases, the baseline deaths using 2015 to 2019 only are slightly lower than the baseline using 2016 to 2019 plus 2021. This slightly reduces excess deaths for 2022 based on the latter data.

John Dee’s Almanac has pointed out that deaths data for England from the ONS in 2022 aren’t yet complete. There’s a huge lack of death at the end of 2022 compared to other years. There could be several reasons for this (delays in processing data, backlogs of autopsy reports…). The Scottish data appears to be more complete, but the data should be re-checked at intervals to find updates.

Excess deaths for all the age groups in the NRS data are shown below. I’ve only shown the 2016-2019 plus 2021 excess deaths data here – the other datasets are available on Github.

I find it hard to see what’s happening in these graphs. The high peaks in April 2020 for most age groups above 55 stick out. Apart from that, it’s hard just to glance at the graphs and see what’s going on. Calculating the cumulative excess deaths for each year can be more helpful.

The cumulative graph helps to show what’s going on more clearly. Looking through each subplot, we can see which age groups had larger excess deaths in 2022 compared to previous years.

Coming back to my original question (Have excess deaths in 2021 and 2022 decreased compared to 2020?) we can look at each graph and check where the bold black line (for 2022) and grey line (for 2021) are higher in week 52 than in 2020 (thin line).

In general, in age groups younger than 50, 2021 and 2022 showed fewer excess deaths than 2020. In some age groups (especially the youngest) there are few deaths in most years, so (for example) a 50% change in cumulative excess deaths may not be very significant.

In other age groups, three show higher excess deaths in 2022 than in 2020. These are 55 to 59, 65 to 69 and 75 to 79. Several show higher excess deaths in 2021 than in 2020. These are 50 to 54, 55 to 59, 60 to 64, 65 to 69 and 70 to 74.

For the majority of the age groups who were considered the most at risk in society, excess deaths in 2021 and 2022 were generally higher than excess deaths in 2020.

Our aim in 2020 was: “Don’t kill granny”. It looks like unexpectedly high numbers of grannies died in 2021 and 2022, despite all our attempts to reduce death and infection.

Update 13/01/22: removed typos in graph subtitles (from “drowingindata” to “drowningindata” (thanks to all who pointed this out!).

What’s the best way to show acute hospital admissions from COVID when all causes hospital admissions are so much higher?

The most recent Public Health Scotland (PHS) statistical report was published a few days ago. One graph that stood out for me was this one on page 45:

At first glance I thought: “Wow, that’s a lot of COVID admissions! Looks like almost half of all acute admissions were due to COVID!”

Then I thought: “Wait a minute, why have they graphed it on a log scale?”

It looks like the authors were faced with a problem. How could they present the changes in COVID admissions while also showing them in the context of all acute admissions?

I tried redrawing the graph from the original data. Unfortunately, daily acute admissions aren’t available as far as I’m aware from the PHS Shiny app (just weekly admissions), the NHS OpenData portal (neither under COVID-19 or Acute Hospital Activity ) or the Scottish Government Daily Dashboard (no admissions data, just the number of people in hospital with recently confirmed COVID-19.)

I used WebPlotDigitizer to extract the data from the plot and then plotted them using R and ggplot2.

First off, just to make sure the data were read correctly, here’s the data plotted on a log scale to match the PHS report.

Graphing on a log scale (with base 10) means that every division on the vertical axis shows a 10x increase on the previous division (you can see that the vertical axis labels go from 10, then to 100, then to 1000.) It’s useful for displaying rapidly increasing quantities or logarithmic units (such as decibels). I’ve never seen it used for comparing levels of hospital admission before.

The advantage of graphing both types of hospital admissions in this way is that it shows the detail in the smaller numbers of COVID admissions. The disadvantage is that it makes COVID admissions look quite large in comparison to the other admissions. The COVID admissions take up half of the height of the plot – so it looks like (at first glance) half of all admissions are due to COVID.

How do the data look when graphed on a linearly increasing vertical scale?

The COVID admissions look a lot smaller now. However, the detail in the rise and fall of COVID admissions is now largely lost.

I prefer to present this kind of data on a faceted graph. This draws individual graphs with different vertical scales and the same horizontal scale. This allows a more detailed comparison of how they both change with time.

This now allows a clearer comparison of the two datasets.

The main features that spring out to me now are:

1. COVID admissions are a small proportion of acute hospital admissions. They range from around 6% (181 admissions) at their peak in January 2021 to less than 0.5% (9 admissions) in May. They also began to rise again in May once the majority of the older people in Scotland were double-vaccinated.
2. Daily all causes acute admissions rose from 2162 at the end of December 2020 to 3267 in early June 2021. This is a 51% increase. This is very difficult to see on the graph with the vertical log scale.
3. The changes in all cause admissions aren’t being driven primarily by COVID admissions. The shape of the all causes graph does not change in the same way as the COVID admissions graph. In fact, all causes acute admissions started to rise as COVID acute admissions began to fall in February – March 2021. Similarly, the rise in admissions in October and November 2021 happened while COVID admissions were falling.

Adding a log scale to the acute hospital admissions graph makes the data more difficult to understand. It has hidden the rise in all cause acute admissions that happened while COVID acute admissions were falling.

Update, 13/01/2022: I asked National Records of Scotland about the dates used. This dataset uses death by date of registration. The “Deaths involving coronavirus (COVID-19) in Scotland – Related Statistics” are based on date of registration unless stated otherwise in the dataset title.

John Dee’s almanac is an interesting substack page written by a former NHS data analyst. They’ve posted several interesting pieces of analysis on COVID data from a variety of UK sources.

One post that caught my attention was this one: Dying with Precision: Causal vs. Co-Morbid Death. This presents a straight line graph of COVID causal versus COVID co-morbid deaths for England and Wales from the Office for National Statistics data.

I have the Scottish data for that period (via the National Records of Scotland). How does that look?

Note the different terms to describe the deaths. In the NRS data they’re labelled “COVID-19 underlying cause deaths” and “COVID-19 contributory factor deaths”.

It looks like the Scottish data isn’t as neat as the English data that John Dee graphs. Note that it’s not clear from the NRS data whether these are deaths by date of occurrence or deaths by reporting date. I’ve asked for clarification on that point (some of the NRS datasets state it explicitly; the one I used here doesn’t). Perhaps that accounts for some part of the straight line behaviour John Dee observes.

The Scottish data still shows a pretty good correlation though – Pearson’s correlation is r = 0.766 (p < 0.001, n = 50, 95% CI [0.724,1.000]).

Interesting!

I was asked recently about Scottish still births data. I graphed some data from the NHS OpenData portal.

There are two sets of data quoted here: the National Records of Scotland (NRS) and the Scottish Morbidity Register 02 (labelled as SMR02 in the OpenData dataset). It’s not clear to me which would be quoted most often in media and Government comments on this – hopefully any reports that use the data will make that clear.

The raw data consists of financial year (from 1997/98 until 2020/21) and birth outcome. Recorded birth outcomes are “Live”, “Still” or “Unknown”.

I’ve graphed the raw NRS and SMR02 data below.

I also calculated the number of still births per 1000 live births for each dataset. This is a standard way of reporting still births.

There’s been an increase for financial year 2020/21. This increase has been enough to trigger investigation from Public Health Scotland as reported by the BBC. I wonder what financial year 2021/22 will bring?

Vaccine passports are being extended in Scotland.

The first minister said the vaccine passport scheme was making “an important and proportionate contribution to stemming transmission”.

https://www.bbc.co.uk/news/uk-scotland-59289707

From the data, the new intervention doesn’t seem to have had any effect on transmission at all since their introduction on 1 October.

Let’s look at the number of positive tests per 1000 people tested across the whole country.

The number of positive tests per 1000 people tested has been increasing since 1 October.

What about changes in cases per age group? Let’s look at positive test results per 100,000 population across eight different age groups.

There’s been a mild increase in cases in 0-14 year olds. In most other age groups they rose near the end of October then started to come back down. Will this be claimed as a positive effect of vaccine passports?

What about test positivity in those age groups?

Test positivity in 20-24 year olds dropped…then increased sharply a few weeks after the introduction of vaccine passports. Test positivity in most other age groups (except the oldest) has continued to rise since October. Are older people benefiting most from vaccine passports?

How did vaccine uptake change?

There’s been a sharp increase in booster vaccines in the oldest age groups. Will this be interpreted as a positive effect of vaccine passports? Is it because of older people who want to get back to clubbing?

Have deaths been affected?

That’s amazing. As soon as vaccine passports were introduced, COVID deaths instantly started to go down. It must have had the most impact on the younger age groups (who were most likely to go to mass events and clubs), yes? Let’s take a look.

Well, no impact on the four youngest age groups. The only apparent effect is on the four older age groups.

Let’s see if a similar amazing effect is seen in hospitalisations:

Looks like the trend is going down…just as vaccine passports are introduced. Success?

In summary, vaccine passports are associated with

• An increase in overall test positivity (transmission isn’t being reduced – the data contradicts the First Minister’s statement at the start of this post)
• An increase in cases in most younger ages
• An increase in booster uptake in older age groups
• A decrease in deaths in older age groups
• A decrease in hospitalisations

This doesn’t mean they caused any improvements in outcome.

The point of this shallow but broad dive into the data is this: if you create a new scheme designed to improve health and you look at enough health indicators, you are more likely to find some sort of data that show it’s working.

Vaccine passports were introduced with no numerical criteria for success. For example, nowhere was it said “we expect vaccine passports to decrease COVID cases by 20%” or “we expect passports to increase vaccine uptake among the young by 5%”. Vaccine passports can be shown to be a success – just define the criteria after they’ve been introduced and look for whatever data supports your belief.

Expect similar graphs from politicians who support the programme. I can even provide a summary text for them to save a bit of time:

Vaccine passports are a helpful intervention in our programme to tackle the pandemic. Already, we have seen their successful introduction has led to a reduction in COVID deaths, fewer hospitalisations and an increase in vaccine uptake. This shows they are a timely and effective method for meeting the greatest challenge our generation has faced.

Scottish politicians who support vaccine passports, soon.

Six weeks after the COVID booster programme started, the data have now been published on the Public Health Scotland OpenData portal. How is it going?

More older people (60+) have taken up the offer of a booster jag, despite the lack of published peer-reviewed data about how the different types of experimental gene therapies interact with each other (there’s a clinical trial which is still running which hasn’t published its results yet, as far as I know). Around 75% of all age groups above 70 years old have received a booster, with lower uptake in younger age groups. Which booster jag is the most popular?

The Pfizer BioNTech vaccination is the most popular booster in Scotland. Sadly, these data do not show which combination of Dose 1/Dose 2/Dose 3 have been administered. There are nine possible combinations:

Having access to this data would be very useful to compare which of the combinations have most effect on mortality and other health outcomes.

Thinking of that, I’ve also updated my graphs on excess mortality (which I first introduced in a previous post) to include the booster. I think of these as a “gobstopper plot”: the more you sook it, the more the colours in the other layers show through. In this case, the more vaccinations are administered, the more colour is seen in the gobstoppers. I’ve also reversed the direction of the fill (yellow is now 0% vaccinated, and blue is 100%) to make it a bit easier to understand. An explanation of the colours and gobstoppers is shown below.

Let’s start with all causes weekly deaths.

Here’s the same again, but focussing on older age groups (60+) from May on.

Next, let’s look at COVID deaths.

Once again, we can focus on the older age groups.

COVID deaths make up 25 to 50% of the excess deaths in most age groups (apart from 75 to 79 – interestingly, it appears that COVID deaths may be the dominant contributor to the most recent excess deaths, although this may be due to the curve fitting method used). Weekly deaths are still higher than we’d expect based on the 2015-2019 data.

It’ll be interesting to see how the situation develops. With the release of the booster data, we can now see how deaths change as the programme continues.

As an aside, I overheard an older lady talking about why she wasn’t getting the booster. “I’m not getting the booster – they’re just experimenting on us older folk. They should stop experimenting on us golden oldies!” Given that none of the data from the UK trials on boosters have been published, wouldn’t it be safer to wait and see what the longer term results of the booster programme from the still ongoing trials are before making a decision on receiving a third dose?

I mentioned previously that I’d like to graph deaths in younger vaccinated age groups (12-15 and 16-17 year olds) alongside the percentage of those populations vaccinated. It turns out that I can’t.

The younger age group categories for vaccination (as defined in the Public Health Scotland (PHS) OpenData vaccination datasets) don’t entirely overlap with either the National Records of Scotland (NRS) historical deaths or the NRS COVID-related deaths datasets. This makes it impossible to create the previous graphs with filled circles showing the percentage of each age group vaccinated. They are fine for 30 and above, but the age groups don’t overlap at all for 0-29 years.

Unhelpfully, the PHS OpenData website only reports excess deaths for “Under 65” and “65 and over” year groups. That’s why I calculated excess deaths based on all cause deaths for all health boards as described in an earlier post.

Instead, I’ve graphed the weekly excess deaths in the younger age groups, along with weekly excess deaths in older groups in a separate chart. I’ve created these as interactive plots on the Plotly website.

The older age groups interactive chart has age groups that overlap with the vaccination data.

The younger age groups interactive chart shows all the age groups for under 30 year olds available in the NRS data.

I’ve also created a plot showing expected deaths and weekly deaths from April 2021 for all the age groups in the NRS deaths data.

If we sum the weekly deaths in each age group from April until now, we get the results shown in the table below.

Scotland has seen 2758 more deaths than expected since April.

Is this okay? Is this an acceptable cost of lockdowns, COVID and vaccination?

I’ve been working on presenting the information that I’ve covered in previous posts more clearly. My aim with those was to show how weekly deaths change as the percentage of each age group vaccinated changes. Now that several age groups are 100% vaccinated, graphing deaths versus the percentage of the population vaccinated doesn’t show us as much.

I’ve now used a splash of colour and some filled circles to show how much of each age group is vaccinated. If I plot these on a graph showing deaths versus time, we can see how weekly deaths are changing in each age group as the percentage of people vaccinated changes.

Our presumption is that age groups that are 100% vaccinated should be experiencing fewer COVID-19 deaths. However, as Norman Fenton points out, a COVID case (and thus a COVID-related death) can mean different things in different places, so looking at the all-cause mortality and comparing that to expected deaths may be helpful to see if the COVID vaccination programme is having harmful effects.

Another thing to think about is that Scotland has seen two large-scale experimental treatments used in the last year: lockdowns and COVID vaccination. The changes in deaths I graph here are affected by these two sets of treatments – with the associated knock-on effects on people’s health of having reduced access to healthcare for several months.

Let’s start by looking at all the age groups that have received most of the vaccinations so far. I’ve yet to add 16-17 and 12-15 year olds to these graphs – soon though, hopefully.

From inspecting the changes in fill colours, it’s clear that most older age groups were 100% vaccinated (two doses) from around May onwards. There’s a lot of variation in weekly death statistics in younger age groups, since so few in those groups die each week. If we filter out the younger age groups and focus on the older age groups (60 and above) from May onwards, we see a clearer picture of how weekly deaths are changing.

Weekly deaths in these age groups are now much higher than expected – around 25% higher in the 80+ age group.

Is this all down to COVID? We can compare the weekly COVID deaths too.

Graphing COVID deaths alongside excess deaths shows that COVID deaths do not account for all of the excess. We should recall that a COVID death doesn’t necessarily mean that a person has died due to COVID. It means they have had a positive COVID test in the previous 28 days before death. That aside, we can’t say that the increase in excess deaths is only due to COVID.

Focussing again on the older age groups (which are 100% vaccinated) shows this effect.

Weekly COVID deaths are increasing, as are excess deaths. It appears that COVID deaths only make up around 50% of the increase in excess deaths. Why is this happening?

If these effects are due to lockdowns and vaccinations, we must ask: is this an acceptable price to pay for the strategy used to manage COVID?

If the excess deaths aren’t due to lockdowns and vaccinations, what is causing them?

Following the end of the school holidays, there has been concern about the number of COVID-19 cases in school-age groups.

Cases have risen very significantly within Scotland and we are looking closely at why that is the case…Undoubtedly the gathering of people together in schools will have fuelled that to some extent, and you can see that in the proportion of younger people who are testing positive.

John Swinney, quoted by the BBC at https://www.bbc.co.uk/news/uk-scotland-scotland-politics-58328945

This has also increased pressure on the Government to push for vaccination of 12-15 year olds and younger to reduce the number of cases and “stop the spread”.

The larger number of cases in younger age groups is clear to see.

The number of daily cases in 0-14 year olds peaked on 30 August at around 1700 new cases. This is almost the same as 25-44 year olds (around 1700 new cases on 30 August) and higher than all other age groups.

The positivity rate is similarly high in comparison to other age groups.

Just under 40% of tests in 0-14 year olds and around 30% of tests in 15 to 19 year olds in the last few weeks have been positive.

What about testing? How many tests are we doing per age group?

We’re testing a comparable number of people per day in the 0-14 years age group as those in the 25 to 44 and 54 to 64 years age groups. How do these compare when we look at the number of tests per 100,000 population in those age groups?

Relatively speaking (by population) we are testing proportionally more younger people than older people. This could explain in part why the absolute number of positive test results is higher in the 0-14 and 15-19 years age groups.

Why is the positivity rate so high for younger age groups? It’s hard to say – in 15 to 19 year olds it’s around 5% higher than in 25 to 44 and 45 to 64 year olds.

One reason it’s hard to say is because the data reported here combines results from the following sources:

• Pillar 1 tests (which are performed regularly on people in hospitals and care homes)
• Pillar 2 tests (which are performed on symptomatic people in the community)

Pillar 2 test results also include the results of confirmatory tests from positive lateral flow tests. Children at school are encouraged to use lateral flow tests twice a week.

This mix of different test results in one set of figures isn’t helpful for showing trends by age group. Confirmatory tests should be reported separately to other tests, and hospital and community results should also be separated.

However, I’ve not found another source for test data that splits up data by age group – if anyone can tell me where I can find

a) the number of daily Pillar 2 tests that are used to confirm positive lateral flow tests results

b) daily Pillar 1 and Pillar 2 test results broken down by age group

I would be very interested to hear from you!

One final note: if vaccination is considered effective at reducing the spread of SARS-CoV-2 in the population, why are around 20% of the tests reported in the last few weeks in the fully vaccinated age groups positive?