[Preprint] Clinical coding of long COVID in English primary care: a federated analysis of 58 million patient records in situ using OpenSAFELY

https://www.medrxiv.org/content/10.1101/2021.05.06.21256755v2

Abstract
Background: Long COVID is a term to describe new or persistent symptoms at least four weeks after onset of acute COVID-19. Clinical codes to describe this phenomenon were released in November 2020 in the UK, but it is not known how these codes have been used in practice. Methods Working on behalf of NHS England, we used OpenSAFELY data encompassing 96% of the English population. We measured the proportion of people with a recorded code for long COVID, overall and by demographic factors, electronic health record software system, and week. We also measured variation in recording amongst practices.

Results: Long COVID was recorded for 23,273 people. Coding was unevenly distributed amongst practices, with 26.7% of practices having not used the codes at all. Regional variation was high, ranging between 20.3 per 100,000 people for East of England (95% confidence interval 19.3-21.4) and 55.6 in London (95% CI 54.1-57.1). The rate was higher amongst women (52.1, 95% CI 51.3-52.9) compared to men (28.1, 95% CI 27.5-28.7), and higher amongst practices using EMIS software (53.7, 95% CI 52.9-54.4) compared to TPP software (20.9, 95% CI 20.3-21.4).

Conclusions: Long COVID coding in primary care is low compared with early reports of long COVID prevalence. This may reflect under-coding, sub-optimal communication of clinical terms, under-diagnosis, a true low prevalence of long COVID diagnosed by clinicians, or a combination of factors. We recommend increased awareness of diagnostic codes, to facilitate research and planning of services; and surveys of clinicians' experiences, to complement ongoing patient surveys.

 

Tweets from the study lead, Ben Goldacre:
https://twitter.com/user/status/1393142399963123712

 

This is exactly as expected, no puzzle here. The most puzzling is in fact that anyone would be puzzled by the expected.

There was no coding until a few weeks ago. A LC coding requires a positive test. Most long haulers did not get a positive test, more than half from surveys and research. Most physicians are not able to recognize LC. Many physicians who may recognize some of the more typical cases are in denial and would not use this coding as they don't believe in it. Most long haulers had multiple experiences of gaslighting and having a mix of anxiety, depression and vague "psychological issues" on their record, there is no chance that medical records accurately reflect reality. Many are simply not going back to see a medical professional because the experience was just that awful, they lost all trust in medicine, at least for the time being, and anyway reality was not recorded at the time and there's nothing they can do to change that.

This is like having a security system where cameras are active but most of them are pointed at a wall and none are recording anyway, it writes straight to /dev/null. Once the event that needs to be analyzed has occurred, it's too late, nothing was recorded. A system has to be turned on in order to work. It was explicitly kept off. So now the whole first year was entirely wasted because no one was actually paying attention and it's an active process, it requires people to pay close attention.

The failure here is one of diagnosis and basic epidemiology. No one was doing the necessary work at the time and there is no looking back in time to fix that because it's something that is either properly recorded or the information is lost forever.

It's the same kind of problem with people arguing that most long haulers don't have a positive test. This is expected for most, finding otherwise is what would be puzzling. This was a product of conscious decisions made at the time, based on assumptions that there was no need for that.

There was clearly a need for that and many pointed it out the whole time. But the medical system is unable to see its own failures and so will rely on its past failures, having not even recorded the most basic data at the time, to fuel its future failures, where it concludes all the wrong things out of the bloody obvious.

I'm sick of "experts" being baffled by the most easily predictable things. It reflects very poorly on the very idea of expertise and the far-reaching consequences of millions having lost all trust in medicine is another one of those failures that require self-reflection to notice, a problem only surmountable by having self-reflection turned on in the first place. This is the odd potent combination of Dunning-Kruger and Murphy's law, and it's just as bad as it sounds.

 

Hi there,

You can leave a comment under the article Please do Go on......

 

Heh why not? I adapted this comment above. Will see if it gets approved.

 

The CBC (national news in Canada), reports everyday on active COVID cases, and those recovered. No mention of a middle ground. No mention of Long COVID in those stats. There have been a few articles on CBC about this, but the focus and advice still seems to be if one gets COVID, then it's 2 weeks to recovery.

How, I wonder can they report those recovery numbers? Is this just extrapolated from the positive diagnoses? Surely there is not that extensive a follow up for all, or even many of those positive cases. Who is doing all those phone calls - thousands per pay? I think not.

 

It's a simple nb_reported_cases - nb_deaths = nb_recovered. Same as everywhere. Zero thought goes into this, you die or you recover, no need to check. Why check on anything that isn't death? If it doesn't kill you it's not interesting. Somehow. That's the idea anyway.

 

You are so right.

 

@rvallee has already hit the nail on the head so I don’t have much to add.

Just that it is commonly accepted that using a “big data” approach will result in all sorts of bias, depending on how that data is collected.

This is mentioned a lot (at least here in the US) in the context of policing; namely that analyzing police data and using it to allocate resources will just result in more discrimination.

I’m unclear if the study authors are intending to make a similar critique, or if they’re actually trying to find insights on long COVID based on clinic-reported data. Reading the author’s tweet thread in good faith, they seem to have some awareness of this issue.

I'm sure that won't stop others from mis-citing the paper however they want to.