Bluesky: https://xitter.cf/t.co/3YRkEkvPrG Linkedin: https://xitter.cf/t.co/6FB8Z97ayu
Norway
Joined May 2013
- Tweets7.3K
- Following481
- Followers1K
- Likes12.8K
Christoffer Baustad Nilsen 🐈 retweeted
The diabetes genetics shenanigans that is going on right now serves as a good example of where many people misunderstand what heritability means.
Fundamentally, when we say something is XX percent genetic or whatever, which is usually from a heritability estimate, we are describing how much of the variation in that trait or disease within a specific population can be attributed to genetic differences.
Nothing about this tells us whether a disease is caused by genetics.
Imagine scurvy, as a simple example. Let's say that scurvy is XX percent heritable because there are genetic differences that influence how susceptible someone is in developing scurvy. That stuff might matter in a population with a low intake of vitamin C, but the disease itself is still caused by low vitamin C, not genes.
Type 2 diabetes is more complex, but the same principle applies. Genetics impact how susceptible someone is to developing it, but they don't cause it. The cause is excessive fat accumulation within the liver and pancreas that drive dysfunction of these organs.
And focusing on genetics isn't helpful anyway because what are you supposed to do about it? It's a purely external locus of control. Even if genetics influence susceptibility, the question always comes back to agency: what can you do to mitigate that risk?
Christoffer Baustad Nilsen 🐈 retweeted
I think Austin and @realDaveFeldman might be somewhat glossing over something pretty crucial here.
Sniderman himself acknowledges there's a lot of uncertainty about which HDL particles, subfractions, or functions might explain these associations.
I also think this uncertainty is **especially relevant** to LMHRs.
As I understand it at least, the LEM predicts larger, cholesterol-rich HDL particles and possibly other changes. But not only do we still not know 100% for sure LEM and LMHR are real things, we also have no idea whether the elevated HDL-C we see in these people is due to more particles, more cholesterol per particle, or better HDL functionality. So technically we don't even know if the biology underlying the HDL in this highly unusual group of people is even comparable to what's underlying the high HDL-C individuals in Sniderman’s paper.
And yet, despite these unknowns they've still written a full substack on it and are sharing about multiple times to their audience, highlighting these dramatic, but modeled apparent risk differences and bringing LMHRs into the discussion at least loosely. Most people reading their stuff are not going to get to the level of nuance and scrutiny I am discussing here, especially if they are of the LMHR phenotype and looking for more reasons to think it's safe.
To their credit, they acknowledge this wasn't an LMHR study in their substack. But they don't really address these potentially very distinct biological differences that could make a big difference on the big picture.
Bottom line for me is many of these so called LMHRs are undoubtedly looking for reassurance about their very high ApoB and therefore might misinterpret this type of stuff as evidence they're protected, when just we don't even know whether the HDL biology potentially underlying Sniderman's findings applies to them.
This feels a bit like this group might be getting ahead of the science once again, although at least with more nuance this time around.
Also to be fair, i think it should be noted that Dayspring's dismissal of HDL-C's independent effects and utility is overstating and possibly not in line with what the available evidence is suggesting.
Expected U.S. spending on AI in 2026 averages about $2,000 per employee, totaling roughly $300 billion a year across all firms.
And firms report modest productivity improvement alongside very limited net effects of AI on employment.
nber.org/papers/w34836
Christoffer Baustad Nilsen 🐈 retweeted
It's actually insane that there are papers coming out about AI capabilities evaluations using two-year-old models.
We have to reduce the length of time to get studies published or academic publishing will have to become completely irrelevant for AI.
When you read a study done after 2050, the probability that the AI has failed to take account of an objection that a human will think of is close to zero.
Christoffer Baustad Nilsen 🐈 retweeted
I have made my first Substack post - Chapter 1 (illustrated) of a Lipid/Lipoprotein Tutorial -- More to follow - Visit thomasdayspring.substack.com… @nationallipid @foundationofnla @society_eas @ASPCardio @escardio @atherosociety @FamilyHeartFdn @TheEndoSociety @fhpatienteurope #cholesterol
Also, on a related note, experts may assume their authority or competence in Field A naturally transfer to authority in Field B, which can trick the expert (and their audience) into believing their opinions carry equal weight in both fields.
Christoffer Baustad Nilsen 🐈 retweeted
I know people are worried about fuel prices. And I get it. You get to the pump, watch the numbers climb, and wonder why it's costing so much more than it used to.
The reality is that global events are pushing up costs, and families here at home are feeling the impact.
But I won't accept that there's nothing I can do to help.
That's why from today, you'll be able to see petrol and diesel prices on Google Maps, compare nearby stations and find the cheapest option before you fill up.
It's a small change, but if there's a practical way to help people keep more money in their pocket, I'll do it.
Christoffer Baustad Nilsen 🐈 retweeted
Can you tell if the ideas in a piece of writing are from a human or an AI?
As AI writing becomes more commonplace, emerging policies on AI use increasingly focus on who came up with the ideas in the text rather than who wrote the text. Modern AI detectors such as Pangram answer the latter question but not the former.
Introducing IdeaLens, a detector of AI ideas!
Christoffer Baustad Nilsen 🐈 retweeted
Statin critics don't understand that absolute risk reduction (ARR) can't be discussed without mentioning the absolute risk itself.
1.2% sounds like a small ARR, until you realise the AR of 1' endpoint in the placebo arm = 2.82% (43% RRR).
ARR can't be > 2.82%.
[Ref: JUPITER]
Hey @grok Based on the best and latest evidence, if I want to avoid a stroke and I can only measure one thing, one metric, what would it be? Concise answer
Christoffer Baustad Nilsen 🐈 retweeted
Actually, events priced at 20% on Polymarket tend to happen around 20% of the time
Christoffer Baustad Nilsen 🐈 retweeted
I think one overlooked aspect of translating genetic estimates of Lp(a)-lowering into expected clinical benefit is the difference in LDL-C and Lp(a) trajectories across the lifespan. The estimates for Lp(a)-lowering are derived by extrapolating from how the effect of genetically lowered LDL-C on cardiovascular risk translates to MACE risk reductions in trials.
LDL-C increases substantially with age, whereas Lp(a) remains largely stable over the lifetime.
If we accept that atherogenesis is driven by cumulative exposure to atherogenic particles over time (“particle-years”), then the effect of Lp(a)-lowering over the same period should be smaller than the effect of LDL-C-lowering. A given reduction in Lp(a) represents a smaller proportion of lifetime cumulative exposure because Lp(a) levels are relatively high earlier in life, whereas LDL-C exposure accumulates progressively with age.
This was not accounted for in the estimates used for the power calculations of Lp(a)HORIZON, which were partially based on genetic estimates derived through this LDL-C → Lp(a) extrapolation.
This could be one reason why the eventual clinical effect of Lp(a)-lowering is smaller than expected.
The Lp(a)HORIZON trial tested one of the most genetically validated targets in medicine, but failed to meet its primary endpoint of cardiovascular event reduction in secondary prevention.
While other therapeutic programs are also supported by genetic data, what was special about the Lp(a) hypothesis was that genetic estimates were used to predict the effect expected in trials, and these predictions partly informed the design of Lp(a)HORIZON.
The results therefore offer a unique opportunity to ask how well estimates from genetic studies translate into clinical risk reduction.
In this new piece (link in the comments), I go through the genetic evidence, the trial results, how they agree or disagree, and where this leaves the Lp(a) hypothesis.
Christoffer Baustad Nilsen 🐈 retweeted
In a recent podcast slide, @realDaveFeldman attempts a defense of the criticism that the LMHR cohort had a high rapid plaque progressor rate of 21% of people with PAV change >1% / yr. (Yellow marking is mine)
(FYI some of cohort participants never had high LDLc at baseline)
"It’s not a frequent topic of discussion, but doctors die, too. And they don’t die like the rest of us. What’s unusual about them is not how much treatment they get compared to most Americans, but how little." pmc.ncbi.nlm.nih.gov/article…
