norwegian covid hospitalization trajectory study
there's something rotten in norway (and it's the design of this study)
sometimes, the most valuable thing you can do for a body of research is help point out which studies within it are without merit and were designed for “waving around” rather that “actually proving things.”
that infamous bangladesh mask study was one such.
to assess this new norwegian hospital study, i brought in a scandinavian consultant:
he had quite a lot to say, none of it laudatory.
much is being made of this “patient trajectory among the hospitalized” study.
unfortunately, most of it is wrong.
the study design and cohort imbalance here makes general comparison deeply fraught to the point where the conclusions foregrounded by the authors look unsupportable.
the key claim is this:
but note that we’re already in trouble from the jump as the size of the active arms here is tiny. 89% of people in the study were unvaxxed. this alone is going to drive all kinds of powering problems and make statistically significant outcomes difficult to attain.
and for most, the effect was tiny.
i have highlighted median days in hospital. for anyone under 80, the gains were negligible in terms of actual days.
and for all people 18+, there was no material difference at all.
but it gets worse.
they’re getting all their risk ratio wins through “adjustments” using their risk model.
all the outcomes they are reporting are a function of massaging the actual data (that fails to diverge) through a model to make it diverge. this is not a valid methodology. we’re just playing computer games and torturing data until it confesses.
so, we’re already well on our way to being able to call this a set of junk conclusions that diverge from raw outcomes only by applying unverifiable risk models vulnerable to GIGO issues.
but it gets worse:
on top of that, they played some slanted games with their definitions.
they did not count the massive risk window of 21 days post dose one as partially vaccinated. it did not even START until dose one +21 days and still counted thru the first 7 of dose 2. whether it was lumped into the unvaxxed to salt that cohort is not clear to me.
right there, they basically blew their results into bayesian smithereens,
but it gets much worse.
these cohorts are, frankly, a train wreck and as so badly matched that it’s impossible to know what’s causing what and there is STRONG reason to suspect that this data going to slant HEAVILY for the vaxxed even if vaccines do nothing.
age is a mess (but favors unvaxxed) “born in norway” is a mess (and favors vaxxed) and underlying risk factors are all over this place. this is why the whole outcome of the study looks like a function of the risk model, not the actual data.
but it gets MUCH worse:
the unvaxxed (left column) were 49% alpha variant and 9% delta. the vaxxed (far right) were 7% alpha and 43% delta. (middle column is partial vaxx)
i mean, come on guys. that alone is a study destroying disaster. alpha had a CFR and hospitalization rate that were on the order of twice delta.
this is because the unvaxxed patients were in hospital much earlier than the vaxxed.
this comparison is not of the same time period.
not only did the variants change, but treatment and knowledge and seasonality did too. this is a literal dumpster fire of statistical confounds that look to heavily slant in favor of the vaccinated.
but here’s where it gets interesting: even with all this rigging, the other outcomes were a disaster for the vaxxed.
let’s look at “admission to ICU” and at “death in hospital.”
in 18-64, the fully vaxxed were more likely to end up in ICU (20% vs 17%) despite a variant half as dangerous. 12.5% died vs 2% unvaxxed. yikes.
there seemed to be ICU benefit from 65-79, but still no death benefit. 12% died vs 10%, itself quite interesting given the lower ICU intake.
over 80 saw no benefit in either.
i’m a little puzzled by the >18 years ICU admission benefit given that unvaxxed did better in every cohort apart from 65-75. something seems off there or perhaps it’s just an intense salting/something of a simpson’s paradox.
but deaths were still hugely variant with 4% unvaxxed and 12% vaxxed. given age variation in cohorts though, i’d be hesitant to make much of a claim on overall and think we do much better looking exclusively at age stratified to get some balance.
the bottom line here is this:
this is a junk study. all it’s headline results come from applying a risk model to raw data that was pretty much at parity.
it has all kinds of confounds including timing and variant.
the error injected here is irretrievable and unfixable and looks to me, especially when segmented by age to remove that issue from cross comparison, to heavily slant toward vaxx.
yet it still shows poor and sometimes disastrous results in ICU and deaths on age balances basis.
anyone trying to tell you this study proves that vaccination shortens hospital stay either cannot read a study or, more likely, did not even try.
this is weak study that, on balance, makes vaccines look more problematic than helpful once you are in hospital, though, frankly, the whole thing is so bad i’d hesitate to draw any really strong conclusions.
also note that this study says nothing about whether vaccines keep you out of the hospital in the first place, itself still a debated issue. i suspect they do on the order of 35-40%, at least for covid (though quite possibly not overall given the side effects) but i also suspect that efficacy is rapidly waning.
there is a lot of bad science being done on covid because, frankly, journals will publish it and health agencies are clamoring for and funding it. so, it’s important we weed out the real detritus.
thanks for shopping: