Is Positron becoming unusably slow/laggy for anyone else after some time? I'm having to restart multiple times a day, e.g., when Ctrl+Enter to send to the console takes multiple seconds. Feels like it's especially happening when working with .qmd files, but I haven't found any clear pattern.
nice paper from @bingkai.bsky.social adding evidence that adjustment in RCTs is good but ML is often not better than linear regression.
arxiv.org/pdf/2602.00434
The result shouldn't surprise you! The signal to noise ratio is too high to learn useful nonlinearities in all but the largest trials.
Very sensible article. This is very clear to anyone working in trials. What will make trials faster? Better outcome measures, designs that allow for frequent interim analyses, analyses that maximize information use.
open.substack.com/pub/cell/p/a...
This paper from Dominic Magirr et al. was very helpful in clearing up most of my confusion: osf.io/preprints/os...
Yes, I think (at least conceptually) I got most of that! I guess I'm currently mostly unsure when I think a superpopulation assumption is useful, and when not. Loved the chapter!
I'm very sympathetic towards Bayesian methods and often use them when I can decide on the approach. But not everyone agrees and imo Markov models are still super cool / promising no matter the framework!
In link.springer.com/book/10.1007..., in the chapter on covariate adjustment @kellyvanlancker.bsky.social clearly states that it's incorrect to use the default delta method, hence my worries.
Thanks. I shall have a look. I meant something similar when I said inference about what happened in our trial sample vs a population in the first post. But maybe that's also not the correct way to think about it? If I just care about "was there an effect in this trial", should I worry about var(X)?
Boostrapping takes too long for simulation studies, so I was hoping simulation from MVN could be good enough, but alas that would ignore the sampling variability of the Xs...
I stumbled across this because we are investigating how to calculate uncertainty for marginal estimands of markov models (a la @f2harrell.bsky.social, onlinelibrary.wiley.com/doi/10.1002/..., but frequentist).
Thanks! I was referring to how marginaleffects calculates the standard errors using the delta method by default, i.e., ignoring the sampling variability. The discussion is very helpful.
I'm slightly confused about this paper: pmc.ncbi.nlm.nih.gov/articles/PMC...
Does this mean that marginaleffects (@vincentab.bsky.social ) underestimates the variance when using avg_comparisons()? Or is this actually a question about whether I want to make inference about my sample vs a superpop ?
The authors don't even report a confidence interval around the difference, but calculating this by hand I get a width of about 35 pp on a scale from 0-100. The uncertainty is just massive. The clinical note completely ignores that and encourages a change in care.
This is obviously a question about practical equivalence or non-inferiority. N = 88. Patients had up to four measurements of TSH, which the authors reduced to 0/1 stable or not for the analysis (surely a crime in @f2harrell.bsky.social eyes). The result is that the study is hopelessly underpowered.
E.g., in a recent issue (clinician.nejm.org/patients-lev...) the clinical takeaway is that a 15% higher dose of levothyrox with breakfast was as good as the standard of care of levothyrox an empty stomach in maintaining stable TSH levels. The underlying RCT is academic.oup.com/jcem/advance...
@nejm.org has a subscription, for clinical notes, which summarize new research for clinicians in bite sized pieces. It's great for e.g., my partner, because she doesn't have time to read studies next to clinical work. But even @nejm.org absence of evidence is confused with evidence of absence...
SWIGs are definitely not core lol
Genuinely seems like a great use of AI? Train a classifier on types of (problematic) changes and map their date to prior to, during, after recruitment/intervention and when I open an article on pubmed/journal the doi/trial-ID is matched and I get a banner with the change info?
I would encourage people to check out the version control of this study on clinicaltrials.gov/study/NCT055....
The definition of early vs late, sample size, and inclusion criteria changed throughout the study period. Once it was even changed to be an observational study and then back to RCT??!
I always find this image a bit misleading because it focus on the year studies are *published*, not when they are *started*.
Here is another version of that figure using the start year of study rather than publication year. Sample sizes in the early 1990s were larger than previous years.
You can of course eat plenty of saturated fat on a vegan diet, e.g. coconut oil and palm oil and that will increase your cholesterol, but a normal-ish plant diet will usually reduce your LDL quite a bit compared to a standard diet.
It's mostly the saturated fat intake (and that also depends on the type of fatty acid, IIRC palmitate is pretty bad compared to stearate for example) that drives the LDL. Dietary Cholesterol does have an effect, is just smaller, reaches a plateau pretty fast.
If you want to go down the rabbit hole, this YouTube channel covers pretty much the whole saga of denialism
youtube.com/@primitivenu...
Back when I was finishing high-school I binge watched all videos, and I remember them as pretty nuanced and honest, surely not perfect but a crazy research effort
"IF statins work (big pharma is not to be trusted) then only because of anti-inflammatory pleiotropic effects not because they lower LDL". ;)
Something something, but Ancel Keys was a fraud, but Minnesota Coronary Experiment, but hunter gatherers, but grass fed beef, but hormones, but LDL is the firetruck not the cause, but ... I might have PTSD from spending too much time debating people on this.
They ref their preregistration but on osf I can't find anything concrete (?). The paper itself doesn't report much you'd expect for an RCT (more details on design, analysis etc).
Anthropic fired their ethicists?
Yeah, not the best rule imo. Figures >> text and tables.
Some journals, e.g., @jamanetworkopen.com don't allow you to only show results in figures. Precise numbers must also be in text. Table + Figure (i.e. duplicating information) is also not allowed. So I guess it often comes down to text + figure?
I think what differentiates this example from statins is that the clinical relevance of AD effects is less clear, so the benefit / risk ratio might not be great. For statins it is as clear as it gets in medicine.
