top of page
Writer's pictureJeffrey Morris

UK death data artifacts: "Stragglers" who delay vaccine doses a select group with higher death risk

The UK Office of National Statistics (ONS) has released all-cause death data for England split out by age and vaccination groups.


Its first release in November included weekly deaths through late September. I wrote a series of blog posts investigating these data, and pointing out that "stragglers" who received their 1st dose on schedule but delayed receiving their 2nd dose had a higher risk of death. They presented 1st and 2nd dose data, and one significant weakness was that it aggregated ages 10-59yr into a single age group, far too wide to be useful.


Its second release in late December included monthly deaths (age-adjusted rates per 100k person-years) through the end of October, and also covered 1st and 2nd dose data, but splitting into finer age groups including 18-39yr, 40-49yr, and 50-59yr as separate age groups. I wrote a blog post plotting and interpreting these data in January, and again reinforcing this "straggler" effect.


On February 4, they posted a third release that includes monthly deaths (age-adjusted rates per 100k person-years) through the end of December. This release also included 3rd dose boosters, splitting the deaths according to the following vaccination groups:

  1. Unvaccinated (no doses given)

  2. <21 days after 1st dose

  3. >21 days after 1st dose

  4. <21 days after 2nd dose

  5. >21 days after 2nd dose

  6. <21 days after 3rd booster dose

  7. >21 days after 3rd booster dose.

The data are are also split into the following age groups: 90yr+, 80-80yr, 70-79yr, 60-69yr, 50-59yr, 40-49yr, and 18-39yr, just as was done for the second release. In this blog post, I plot and try to interpret aspects of these data. My plots are done as follows:

  • I plot the monthly death rates (age-adjusted deaths per 100k person-years) along with the 95% bounds to indicate level of uncertainty (wide bands correspond to small sample sizes and high levels of uncertainty).

  • I make the width of the lines proportional to the number of individuals in that vaccination group at that time, so that one can see that when a line is thin there are very small numbers of individuals in that vaccination subgroup at that time, while times for which the line is thick indicates a high proportion of the age group in that vaccination subgroup.

  • For each age group, I do 3 sets of plots, one for each dose (1st, 2nd, 3rd booster dose): each contains the unvaccinated group (green) as well as the <21d (blue) and >21d (red) groups for the respective dose.

Here are the R scripts that download the data from the UK website and generate all of these plots (you will have to change the setwd() command to match your local directory in order to save the plots):

As I have previously observed, some of the key features in these death rate curves occur at specific times in the vaccine rollouts. The UK posts data on daily number receiving 1st, 2nd or 3rd dose vaccines daily by age groups. Here is the .xls data downloaded from this website as of February 3, 2022:

These data are not in a user-friendly format at all, but I have written an R script to parse out the relevant data, organize it into the same age groups as the UK ONS death data, and to plot the proportion in each vaccination status for each age group over time to characterize the specific timing of the age-specific vaccine rollouts:


For this blog post, I will highlight one specific artifact that is persistent in these data in which the small set of "stragglers" who delay receiving either 1st, 2nd or 3rd dose have higher death rates than those receiving their doses on schedule. I focus on all-cause deaths here.


First, I plot the 90yr+ data, stacking the plots for the 1st, 2nd and 3rd doses along with the corresponding vaccination plots. The vertical green, red, and blue lines correspond to the time point at which the vast majority in the 90yr+ age group have received 1st, 2nd, or 3rd booster doses, respectively.

The bottom panel shows the vaccination of the 90+ age group, demonstrating percent of age group in each vaccination status.

  • By late January, the vast majority of the age group had received 1st dose of vaccine, and the small proportion left unvaccinated, the "stragglers" who had not vaccinated on schedule had ~3x the death risk of those who received 1st dose. This is evident in the unvaccinated group (green line, top plot).

  • By mid-April, the vast majority of those in the age group who had received a 1st dose had received their second dose, and the "stragglers" who did not receive their 2nd dose on schedule had >3x the death risk of those who received 2nd dose on schedule. This is evident in the group >21d after 1st dose (red line, top plot)

  • By early November, the vast majority of those in the age group who had received a 2nd dose had received their 3rd booster dose, and the "stragglers" who did not receive a 3rd booster dose on schedule had 2.5-3x the death risk of those who received 3rd booster dose on schedule. This is evident in the group >21d after 2nd dose (red line, second plot).

While the death rates are very high for these vaccination subgroups at these specific time points, the corresponding death counts are not so high since these spikes of death rates occur when the corresponding vaccination subgroup gets very small, select group comprising a small percentage of the age group.


We see this pattern for all the older and middle age groups.


Here is 80-89yr, where the "stragglers" at all 3 dose groups had 5x the death rate of those who received their doses on schedule:


Here is 70-79yr, where the "stragglers" at all 3 dose groups also had 5x the death rate of those who received their doses on schedule:

Here is 60-69yr, where the "stragglers" at all 3 dose groups had 3-5x the death rate of those who received their doses on schedule:

Here is 50-59yr, where the "stragglers" after 1st and 2nd doses have death rates substantially higher than those who received their doses on schedule. The 3rd dose boosters were just taking off late December, but we see the death rate of those >21d after 2nd dose increase as it becomes a more select group (<25% of age group). We expect this group's death rate to increase once data for January are released.


What could be causing this phenomenon?

  1. Missclassification: Neil et al. suggest that it could be the result of misclassification -- that deaths soon after 1st dose rollout were misattributed to unvaccinated, deaths soon after 2nd dose rollout are attributed to 1st dose, and deaths soon after 3rd dose booster rollout are attributed to 2nd dose. It is difficult to envision how that could be the case given the ONS specifically defined categories <21d after 1st, 2nd or 3rd dose. It is clear from these definitions that they did not routinely assign events the first week or two after inoculation to the unvaccinated group, as some claim without evidence. I have yet to seen a reasonable explanation for how one proposes the misclassification error occurred, or how it would explain this artifact, but given the timing it is possible there is some sort of misclassification or misrecording of some of the death times.

  2. Selection bias: We consistently see this increased death risk at the point when the vast majority have received their next dose, leaving a small and shrinking select group not receiving their dose on schedule. This small group of "stragglers" includes those too sick to receive vaccine, who would clearly be a select group at higher risk of death. It would also include those who experienced medical complications after the previous dose, and so delayed or refused subsequent doses, another group one would expect to have a higher risk of death. We need more information on the demographic characteristics of these "stragglers" to see if there are any evident factors explaining the higher death risk.

Given this undeniable pattern in the data, it is important for the UK ONS to investigate and understand it. The death dates should be carefully checked to ensure there is no misclassification, and the relative demographic and clinical characteristics of the "stragglers" delaying 1st, 2nd or 3rd dose relative to their counterparts receiving the recommended doses on schedule.


It is always tricky to interpret observational data, but this is especially true in a dynamic setting like the vaccine rollout in which the vaccination prioritization and individual vaccination decisions impact the cross-section in each vaccination subgroup, with some of these subgroups becoming small, select subgroups with higher death risks. Follow up studies to characterize these subgroups is necessary to understand the patterns in these data.


For some big picture perspective, here are the age-adjusted death rates for each vaccination group aggregated over the entire year:

We see that overall, the age-adjusted all-cause death rates are substantially lower in the vaccinated groups than the unvaccinated. The artifacts discussed above only involve small, select subsets of individuals, and are strongly outweighed by the lower death rates for the same vaccination subgroups during times in which they contained much larger numbers of individuals.

3,316 views204 comments

Recent Posts

See All

204 Comments


markmrook
Apr 01, 2022

The following pre-print found a higher rate of COVID infection in the first 14 days after the first dose, a period that is generally excluded in other vaccine effectiveness studies:


Figure 2 shows comparison of time to occurrence of event (COVID-19) between vaccinated and unvaccinated groups. Individuals belonging to ‘1’ dose group developed COVID-19 earlier compared to the unvaccinated and ‘2’ dose group with statistical significance. After adjusting for potential confounders in Cox proportional hazard model, age, sex, prior history of COVID-19 and vaccination status emerged as tentative determinants of occurrence of COVID- 19 (Table 1a). The risk of occurrence of COVID-19 was nearly 1.5-times for those <40 years of age as compared to participants ≥ 40 years, 1.2-times higher…


Like

markmrook
Mar 11, 2022

This from the Lancet is interesting:


Estimating excess mortality due to the COVID-19 pandemic: a systematic analysis of COVID-19-related mortality, 2020–21

https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(21)02796-3/fulltext

Like

Alex
Mar 09, 2022

https://www.theblaze.com/news/denominatorgate-how-public-health-agencies-are-skewing-the-statistics-on-vaccine-effectiveness


Hi there - I am curious your opinion on this article about skewed vaccine efficacy. There will always be error in human data collection, however some of these estimates seem awfully large!

Like
Jeffrey Morris
Jeffrey Morris
Mar 09, 2022
Replying to

Yes — that is why these population data themselves are not sufficient for estimating VE — you have to adjust for confounders and also compute VE as a function of time since vaccination to adjust for the decrease of protection from infection from the reduction of circulating antibodies and greater dependence on other elements of the immune system like B cells producing new antibodies that take longer and are often not fast enough to prevent positive PCR tests. But fortunately there are some good studies out there that do adjust for these factors — most are peer reviewed papers, but often it is the hot takes involving summaries of raw data on social media that get greater attention.

Like

markmrook
Feb 17, 2022

I decided to take a look at monthly excess death data for England in 2020 and 2021 based on the UK ONS data. The original ONS data is age-standardized and seasonally adjusted, so these shouldn't be factors. I made some adjustments to increase meaningfulness, which I describe below at the end.


I first plotted excess monthly deaths in 2020-2021 vs. 2019 for the following causes: all causes, COVID-attributed, Ischaemic heart disease, Cerebrovascular disease, Dementia and Alzheimer's disease, and "Symptoms, signs and ill-defined conditions" (which is their catch-all for "unknown cause").



Because the total excess deaths and COVID deaths were so much greater than the other causes, I rescaled these per 20,000 people instead of per 100,000 people, to better show…




Like

markmrook
Feb 17, 2022

Xie et al., who published the alarming article regarding increased cardiac risk after COVID infection, have now published a similar article in BMJ regarding increased mental health outcomes after COVID infection:


Risks of mental health outcomes in people with covid-19: cohort study

BMJ2022; 376 doi: https://doi.org/10.1136/bmj-2021-068993(Published 16 February 2022) Cite this as: BMJ 2022;376:e068993


https://www.bmj.com/content/376/bmj-2021-068993#aff-1


However, it might not be as scary as it seems, as pointed out in an accompanying editorial:


https://www.bmj.com/content/376/bmj.o326

Like
Post: Blog2_Post
bottom of page