Friday, 14 January 2022

More statistics

What follows being kicked of by a puff for the paper at reference 3 in an email from Medscape at reference 1. And having lost the email, it was easy enough to recover the paper from PLOS at reference 2. The work of researchers from the UK, from Norway and from Russia.

This work look at the excess deaths among a large cohort of people, near 10 million older adults, living in the UK, during the first peak of the Covid-19 pandemic, that is to say between March and May 2020.

The method was to side-step the problem of whether any particular death was due to Covid-19 or whether it was due to something else, by looking instead for groups of people among whom there were more deaths in total than you might have expected on the basis of what had happened in those groups since 2015, in short for groups of people experiencing excess deaths. So, for example, they looked at the experience of people with rheumatoid arthritis. The basic data were the weekly death counts. With denominators in million person weeks.

For the sample population as a whole there were 35,000 deaths in the relevant period, 400 per million person weeks, of which 130 or 48% were excess over expected. Near 20,000 of these deaths were people aged 80 and over and something over 8,000 were in the 70-79 range. Men and women roughly equally. While, for example, 1,000 of them were South Asians (98% excess) and just over 1,000 of them had rheumatoid arthritis (48% excess). While, oddly, just 370 of them were smokers with the much smaller excess of  20%. Ex-smokers fared rather worse, but perhaps they tended to be older.

The headline result is that while it is true that if one was old or ill, yes one was more likely to die, but, relatively speaking, the arrival of Covid-19, did not make it any more likely, at least not during this first wave, not in the circumstances of this first wave. Yes, you were more likely to die, but the increase in likelihood, the increase in risk, was much the same as for everybody else. Or in the words of the paper: ‘… this suggests that COVID-19 has dialled up the risk of death by a similar proportional degree for most people…’.

But there were four exceptional groups: people with dementia, people with learning difficulties, people of colour and people living in London. The first two of these groups were more likely to be living in care homes and the last two were more likely to be living in crowded homes or to be in public facing occupations, for example retail, health, care or transport.

Note that this work tells us what the mortality outcome was. But while it is suggestive that the four exceptional groups were more likely to get infected, it does not tell us whether a greater prevalence of infection was compounded by a greater likelihood of infections being serious. Finding out about that is another matter altogether.

I was impressed by the fact that the researchers were able to draw on a very large sample of people drawn from GP records. This sample comes from Clinical Practice Research Datalink (CPRD, to be found at reference 4) which collects anonymised patient data from a network of GP practices across the UK. Primary care data are linked to a range of other health related data to provide a longitudinal, representative UK population health dataset. The data encompass 60 million patients, including 16 million currently registered patients. This very powerful resource is made available to appropriately qualified researchers and research.

I was interested to see that psoriasis, an obscure skin complaint which I suffer from, was among the health conditions looked at. Maybe it is not so obscure after all.

On the other hand, I did not understand the detail of the statistical work, which appeared to be involve estimating relative rates of death (RRs) using something called seasonally adjusted negative binomial regression models. 

As far as I can remember and as far as I can make out, such models are used when one knows how often something happens on average, in this case whether someone in some group or other is going to die or not in the course of a week, and one wants to know the distribution of the total number of deaths of people in that group in the course of some particular period, in this case close to three months.

While seasonable adjustment, all the thing with government statisticians when I was starting out in the 1970’s, was not something I was ever very comfortable with.

So while it all looks reasonable, I have to take the results on trust. And hope that my trust is not misplaced.

And that our politicians are building their trust by gently poking things around; poking the results and the researchers. Taking second opinions. Having good support from their advisors and civil servants. Given that there are only so many hours in the day, what else can they do?

References

Reference 1: https://www.medscape.com/today

Reference 2: https://plos.org/

Reference 3: Factors associated with excess all-cause mortality in the first wave of the COVID-19 pandemic in the UK: A time series analysis using the Clinical Practice Research Datalink – Helen Strongman, Helena Carreira, Bianca L. De Stavola, Krishnan Bhaskaran, David A. Leon – 2022.

Reference 4: https://www.cprd.com/

No comments:

Post a Comment