Corey Scholes

If you could switch off your surveillance on my calls with clinicians who refuse to do anything until theyve "seen the data"...that would be great

Is it possible to know if the patient is a candidate for surgery at the start of prophylaxis? Or is everyone staryed on prophylaxis regardless of surgical appropriateness and then referred on failure? If its the former, I think that would simplify the model a fair bit

Do you have any consistency in the definition of failure of prophylaxis? These may differ ir be applied differently between referrer and surgeon

Just trying to get my head around the context here...i assume the drs making the referral are primary care and then a surgeon is reviewing? So is your endpoint the rx to surgery or the surgery itself? Depending on the condition, not sure if elective or otherwise, it may not be 1:1 from rx to sx

Horseshoe crab

Its the horse crab of universal science assets...i dont know how, bit its surviving an extinction level event..

Is this how our elderly citizens feel when youngsters ask them about historical events - they didnt even realise they were witness/participant at the time? Thats how i feel about the statstwitter moments...i remember them, but didnt appreciate their significance until now....classic statstwitter..

Old mate went truly all in on in that round

Fraud in army - bold strategy at any time
Fraud as an Other Rank
Crime in early 20th century (what is rehabilitation?)
Combine all of those things and then do it in wartime...

"Model failed to converge"

#rstats #medsky #statsky

How to Train Yourself to Be More Open-Minded and Less Easily Misled New research shows that cultivating open-minded thinking can be taught, and that this simple shift improves our ability to tell facts from falsehoods.

Can we actually train ourselves to be more open-minded and harder to fool?

New research says yes. A short message warning us about closed-mindedness made people better at spotting misinformation, less likely to believe conspiracy theories, and more thoughtful about what they shared #MisinfoResearch

Interception reduction from deforestation and forest fire increases large-scale fluvial flooding risk - Communications Earth & Environment Forest coverage loss from wildfires and deforestation increases flood risk for large-scale river catchments by reducing interception, according to analysis of data from twenty-one watersheds in Australia.

🌧️ Forest loss and flooding

A new study shows that deforestation and forest fires in Australia reduce rainfall interception, increasing river flooding across vast regions.

🔗 www.nature.com/articles/s43...

#SciComm #ClimateCrisis #Floods 🧪

I think Moriarity had had some particularly bad experiences with German tanks in the bocage though

This is the way

First episode our hero (or Xena-esque heroine) comes up against an ortho bro|sis cartel pushing out some dubious "trials" on their new pet surgical techniques that they are introducing in young athletes. There are strong echoes of the Terminal List when one of the athletes joins in the hunt.

Kicking in doors on hospital Research Offices and pushing the envelope on kidnapping and false imprisonment to hrec chairs that have allowed dubious methods through the approval process.

I think a Reacher + The Accountant remake is needed - a drifter statistician with an outrageous physique moving from one random hospital to the next, breaking up clinical research cartels and applying high levels of violence to restore appropriate rigour to studies in their early phases of set-up.

I spend an inordinate amount of time explaining the relationship between inc-exc criteria and the recruitment flowchart - "yes but who are you excluding AFTER you initially included them?"

the strictness of forcats upfront saves so much hassle downstream...nothing worse than untangling factors several models down the way

I should add some tags so it comes up on feeds

#rstats #statsky #clinicalresearch #whatislove

The options I see are - run everything with a subset + impute approach, ignoring the advantages of the reverse approach; use complete case analysis for the flowchart and descriptives (risking a disconnect with the model dataset and results) or there is a way of working with mids for every step?

The purpose of imputation is to input into the inferential modeling step (what effect does x variable have on y outcome?) but if the imputation is upstream of the sample selection I need to figure out how to work with a mids object for generating descriptive summaries + the recruitment flow chart

For background I am aligning to the RECORD reporting guidelines in a quarto document within Rstudio. This means if I move the imputation step from where it is in the pipeline currently (RECORD 12.2 data cleaning) to upstream (RECORD 6.1 sample selection) - I am going to have some...complications

Thanks to everyone that weighed in on this - there are some strong arguments for imputing first and then subsetting, especially when the subsetting variable set contains missingness.

But I am wrestling with how to implement this across the entire analysis pipeline in practice.

Dammit, you beat me to the reply

Starting a thread like this with this level of heresy needs to come with a warning label

I think this is a separate question to the one I'm asking for this analysis, but would be open to collaborating with others on a more methods paper

ah ha there is always a paper...practically there is missingness in the subset variables (although it could be argued away), so it looks like impute-then-subset.

The major downside being how complicated this analysis file is going to become!

Thanks for weighing in!

subsetting based on patient characteristics

Ok #skyhivemind and #statsky #rstats - I am analysing a subset from a clinical quality registry. Both the total registry dataset and the subset have missing data that needs to be imputed. Should I i) impute the full registry set and then subset or ii) subset from the get-go and discard the rest?