Donald Szlosek
@dszlosek.bsky.social
Biostatistician @IDEXX formerly at harvardmed, @BIDMChealth, @nasa. Big data, clinical trials, and medical diagnostics. Mainer. Opinions are my own. he/him
Interesting read! In spirit it reminds me of the Vibration of Effects work by Patel (2015) and colleagues, although their work was solely focused on analytical choices.
November 16, 2025 at 3:11 PM
Interesting read! In spirit it reminds me of the Vibration of Effects work by Patel (2015) and colleagues, although their work was solely focused on analytical choices.
My knee jerk reaction is to almost always go with clopper-pearson (exact) mostly because the coverage is so good near 1 and 0 (or at least that is what I remember reading in a paper many years ago).
November 16, 2025 at 2:48 PM
My knee jerk reaction is to almost always go with clopper-pearson (exact) mostly because the coverage is so good near 1 and 0 (or at least that is what I remember reading in a paper many years ago).
I was recently working with a distraught med student who told me one of his classmates had 75 (!!) publications!
November 12, 2025 at 11:00 AM
I was recently working with a distraught med student who told me one of his classmates had 75 (!!) publications!
I have wondered about this exact thought! Superb. Also love the Genstat output
November 11, 2025 at 1:43 PM
I have wondered about this exact thought! Superb. Also love the Genstat output
my chest hurts reading this.
November 11, 2025 at 1:20 PM
my chest hurts reading this.
I would be very curious to hear @maartenvsmeden.bsky.social thoughts on handling large volume on prognostic model comparison. Single database? Just smaller time window? I think I remember some justification of think in a PROSPERO doc of yours.
November 8, 2025 at 7:24 PM
I would be very curious to hear @maartenvsmeden.bsky.social thoughts on handling large volume on prognostic model comparison. Single database? Just smaller time window? I think I remember some justification of think in a PROSPERO doc of yours.
Exactly my thought re: multi-state model and msm package I mentioned earlier.
November 8, 2025 at 2:27 PM
Exactly my thought re: multi-state model and msm package I mentioned earlier.
Is aging out a competing risk? Outside of LTMLE, multistate models (MSM package in R) + IPCW would work well here if I am understanding the question correctly.
November 8, 2025 at 12:53 PM
Is aging out a competing risk? Outside of LTMLE, multistate models (MSM package in R) + IPCW would work well here if I am understanding the question correctly.
If they check back every month to assess complexity (cpx) you could treat cpx as a time varying covariate and run something like Longitudinal Targeted Maximum Likelihood Estimation. You would essentially IPW at each check in to reweight cpx. Time to event would to time to surgery ~ defer time.
November 8, 2025 at 12:49 PM
If they check back every month to assess complexity (cpx) you could treat cpx as a time varying covariate and run something like Longitudinal Targeted Maximum Likelihood Estimation. You would essentially IPW at each check in to reweight cpx. Time to event would to time to surgery ~ defer time.
because their date of death has not yet been accurately recorded or linked to the study database.
November 7, 2025 at 6:18 PM
because their date of death has not yet been accurately recorded or linked to the study database.
GT bias arises from imperfect/delayed ascertainment of mortality (death) data. It happens when researchers inadvertently consider patients to be "at risk" (still alive and in the study) for a longer period than they actually are.
November 7, 2025 at 6:18 PM
GT bias arises from imperfect/delayed ascertainment of mortality (death) data. It happens when researchers inadvertently consider patients to be "at risk" (still alive and in the study) for a longer period than they actually are.
Every model comparison paper that never discusses the details of variable selection and transformation😤
November 7, 2025 at 11:42 AM
Every model comparison paper that never discusses the details of variable selection and transformation😤
If people are using the correct statistics, casual inference and study design, they will quickly realize they don't have the funds to design the study, the time to learn the statistics, or the right data to answer the actual question they are interested in. Then they will never publish or get tenure
November 7, 2025 at 10:44 AM
If people are using the correct statistics, casual inference and study design, they will quickly realize they don't have the funds to design the study, the time to learn the statistics, or the right data to answer the actual question they are interested in. Then they will never publish or get tenure