Thread by @dgurdasani1

A thread slightly borne out of frustration on the widely misrepresented discourse on long COVID, esp in children. This is for the 'long COVID studies in children with controls are rigorous and loads of controls have symptoms so not sure this syndrome is real or important' group🧵

First, controls per se *do not* make a study sound - How do you account for the fact that children are often asymptomatic acutely, serorevert quickly or don't seroconvert at all. Long COVID itself is associated with lower Ab levels, furthering this bias.

https://twitter.com/dgurdasani1/status/1439905644409245696?s=20

Rigorous science is v. important, but let's not pretend that studies are rigorous because they have controls (just like people pretend RCTs are superior to observational evidence by virtue of being RCTs, even if they're conducted badly).
pubmed.ncbi.nlm.nih.gov/34273064/

When control groups are contaminated with cases (people who were infected but either didn't get tested or tested antibody negative), it leads to misclassification, which will underestimate prevalence. This bias is made worse by long COVID itself being associated with lower Abs.

Many long COVID studies haven't included key symptoms identified in surveys (e.g. brain fog, mood changes in the Zoe app study). And whether questions about specific symptoms were asked systematically vs open questions also makes a bit difference- anyone doing surveys knows this.

When follow up itself is biased e.g. in the Zoe app study, where reporting to indicate resolution of symptoms was much lower in cases than controls. And more incomplete follow-up in the group that was still reporting symptoms compared to those where symptoms resolved early.

This means it was likely that those who didn't follow up perhaps didn't because they were looking after sick children and had other responsibilities. In this case the fact of non-reporting in itself can be associated with illness and should not be used to indicate resolution

We considered different numbers of those not following up as having long COVID and found this could make very large differences to the estimates, if loss to follow up was actually dependent on having longer-term symptoms.

Accounting for waxing and waning of symptoms is also important, as this seems to be one of the key features of long COVID in patient surveys. Studies that don't allow for this will underestimate it.

Using 'any' symptom definitions alone is also problematic, rather than combinations of symptoms, and how they change over time, and proximity to previous infection - which are all important. There are very clear patterns of symptoms in long COVID. These need to be considered.

e.g. pre-menstrual syndrome is a constellation of symptoms along with timing (near the menstrual cycle). If one measured the symptoms in the general population, symptoms like bloating, constipation, breast tenderness are quite common. The pattern and when it occurs is important.

It's no coincidence that studies that look at combinations of symptoms find these are far better at spearating cases from controls. in long COVID cohorts. Multiple long COVID surveys show that clusters of symptoms exist & are similar. Here from the CloCK study:

Our work, CLoCK, and REACT-1 uncovered similar clusters & risk factors. We uncovered a respiratory system + fatigue cluster (fewer symptoms) and a far more multi-system symptom cluster with prominent neurological symptoms.

Women, those with pre-existing conditions, socio-economically disadvantaged, & those with multiple acute symptoms were at greater risk. This is very consistent with studies like REACT-1 and CLoCK carried out in different groups.

Clusters correlated strongly with functional indicators around day to day activities being impacted, sick leave, ability to care for others, suggesting they were likely functionally significant. These are reproducible and consistent patterns across patients reporting long COVID.

Several surveys, including ours, examining patterns of symptoms clearly show specific patterns in different studies- e.g. neurological symptoms often become more prominent over time, and are long lasting. Fatigue is long lasting.

Systemic symptoms like fever, chills and loss of smell & taste are less persistent.

So controls are important - but
-they need to be defined well - not just on cross-sectional serology, or negative tests at a point in time (given high asymptomatic infection, esp in children)

-constellations and clusters need to be compared between cases and controls too. 'Any' symptom based definitions don't really capture syndromes
-longitudinal patterns of symptoms also very important to consider and compare between groups.

'Any' symptoms being common in a control group is unsurprising, and does not 'rule out' a syndrome or tell us about it's prevalence. You can think of many well-known clinical syndromes with well-understood underlying biological mechanisms, and you'll realise this is true.

So, let's not pretend that long COVID studies with controls are somehow not limited or biased. They absolutely are. If you want to do long COVID research justice, involve patients, look at patient surveys and patient led research to understand patterns, and then study them.

Once you understand constellations of symptoms and patterns (yes, this actually requires surveys and studies that *don't* involve controls at first - just as we do observational case series initially in medicine to figure out what's going on, before we systematically measure it!)

Once you understand the possible constellation of symptoms and patterns, you can then study them in controlled studies- but do study the constellations and patterns, and not just studying 'any symptom' and concluding that it's 'all in the mind' or 'lockdown fatigue'.

Also, don't forget that there are now multiple studies showing biological correlates of long COVID- including neurodegeneration (yes, these too had controls, and pre- and post-covid scans!), autoimmune markers, neuro-vascular disease, renal disease, even in 'mild cases'.

So perhaps try to build a comprehensive picture of the syndrome and the biology rather than limited flawed analyses of biased studies with flawed controls that make little sense from a medical or clinical perspective.

Our critique of the Zoe app study here outlining the biases when loss to follow up may be associated with the outcome. Excluding missing data and re-analysing (the sensitivity analyses described by the authors) does not address this systematic bias:
thelancet.com/journals/lanch…

And to those doing systematic reviews that say much of the same- unless you're critically evaluating the above biases, and addressing them, those reviews aren't worth much, and don't do justice to the people suffering with long COVID who deserve more critical & rigorous research.

One last bit- having listened to voices from the long COVID community, I can say that the worst part of dealing with flawed long COVID research has been watching people with long COVID being re-traumatised and gaslit again & again by badly conducted studies.

I don't think people realise the trauma of having a severe debilitating condition that affects your day to day life, and then having flawed research constantly quoted to tell you that your suffering isn't real. It's up to scientists to do much much better by patients.

Here are some links to studies that show a lot of consistency in patterns and predictors for those interested - like one would expect to see for a... syndrome!
medrxiv.org/content/10.110…
spiral.imperial.ac.uk/bitstream/1004…

thelancet.com/journals/eclin…

researchsquare.com/article/rs-798…

I'd really like to move the discourse from 'does it exist?' to 'how is it impacting people long-term, what are the biological correlates, how do we prevent and treat it?'

Sorry, apparently, the REACT study link doesn't work for some - you can use this link and navigate to the June long covid react study and download the paper:
imperial.ac.uk/medicine/resea…