Biostatistician/Professor/Founding Chair of Biostatistics, Vanderbilt U. Blog: Statistical Thinking:fharrell.com @f2harrell on bsky.social
Joined January 2017
- Tweets 19,559
- Following 175
- Followers 30,202
- Likes 34,685
432 Photos and videos
Frank Harrell retweeted
Jun 12
Determining the Physiological Threshold for Angina (ORBITA-FIRE): A Double-Blind, Randomized, Placebo-Controlled Study.
doi.org/10.1161/circulationa…
Contributions from Baim faculty, @f2harrell and @rallamee.
#BaimInstitute #ClinicalResearch
2
4
785
Frank Harrell retweeted
Jun 9
This one hits hard.
1,751
16,565
74,989
1,174,182
Frank Harrell retweeted
“Turning loose” ML on high dimensional observational datasets is highly problematic, see hbiostat.org/bbr/hdata, @f2harrell has nice clear examples and explanations. His work is *must read* for MDs, PhDs, etc
1
1
23
2,559
Frank Harrell retweeted
This is the bias you cannot sample size your way out of. Power calculations assume SE = σ/√n → 0 as n → ∞. But if E[θ̂] = θ δ, then MSE = Var δ².
1
2
10
1,854
Frank Harrell retweeted
This paper claims an AI test predicts who benefits from abiraterone in prostate cancer. Interaction p=0.02. Everyone is excited. I read the methods. You should too. Thread.🧵
annalsofoncology.org/article…
1
4
30
11,538
Frank Harrell retweeted
May 31
This review is a helpful entry point into TT thinking, but it may overstate the extent to which TTE can support causal inference when applied to retrospective data. TTE can help avoid preventable design errors, including self-inflicted biases. It does not recreate randomization, blinding or prospective capture of the factors that drive both tx choice and outcome. For that reason, many CV TTE studies should be described more cautiously: as structured observational comparisons, or at best hypothesis-generating, unless the data source was prospectively designed, audited, and shown to contain the main determinants of tx choice and outcome. As @f2harrell has argued, a more accurate label for many of these studies may be “Best Practices in Observational Comparisons” rather than TTE.
2
2
17
2,421
May 28
New video presentation on the FDA Bayesian guidance: fharrell.com/talk/bguide/ #Statistics #Bayes #pharma #rct #clinicaltrials @fda @US_FDA
18
56
3,345
May 25
New presentation on my experiences and thoughts about the roles for AI in #Statistics: fharrell.com/talk/ai #LLM #ai
4
20
73
5,676
Frank Harrell retweeted
May 21
17th Annual University of Pennsylvania Conference on statistical issues in clinical trials - Covariate adjustment in randomized clinical trials.
doi.org/10.1177/174077452614…
Contributions from Baim faculty, @f2harrell.
#BaimInstitute #ClinicalResearch
2
7
1,028
May 20
#Statistics thought of the day: Virtually all of single-event survival analysis (including Kaplan-Meier, Cox PH, log-rank, AFT) is a special case of cumulative probability semiparametric ordinal regression models: hbiostat.org/rmsc/ordsurv
1
6
47
3,412
May 20
For those interested in behind-the-scenes statistical theory: The log-log link is the only link for which a Cox partial likelihood is possible. Ordinal models use full likelihood so handle all links, and estimate the underlying survival curve S0(t) simultaneously with beta.
3
14
1,044
May 16
Computing tool of the day: I've needed to have an elegant way to stitch together pieces of Quarto books to make a unified html file for teaching. In collaboration with Claude I now have a great yaml-driven tool for this. #rstats
1
3
46
3,867
May 17
What makes the resulting html file suitable as a slide deck: Using a chrome-based browser such as Brave so that math and text have correct font sizes upon 200% zoom. Quarto html automatically removes content in margins and adjusts line breaks when zooming.
1
1
766
May 13
The ORBITA group continues to do groundbreaking, innovative cardiovascular research, and questioning commonly held beliefs in cardiovascular medicine. Proud to be associated with them.
May 12
ORBITA-FIRE suggests universal Ischemia-based thresholds for FFR and non-hyperemic pressure ratio should be interpreted with caution: Integrating symptom-linked physiology may refine PCI selection and improve symptomatic response.ahajrnls.org/3R6AzQD
2
12
70
19,854
Frank Harrell retweeted
50 years of oncogene theory. One broken premise: that we can identify the causal driver, that it’s singular, that inhibiting it collapses the network. We can’t. It isn’t. It doesn’t. The oncogene is not the disease. It’s the mutation we happened to sequence.
May 7
For 30 years, we couldn't touch RAS.
It drives 90% of pancreatic cancers.
A new drug in today's @NEJM just proved we can.
Early data. Real responses. More to come.
@OncoAlert @DanaFarber
8
3
44
16,397
Frank Harrell retweeted
@kalashjain1 Hello, you can read it here: threadreaderapp.com/thread/2… See you soon. 🤖
1
2
920
Frank Harrell retweeted
🧵 A new Nature paper on “spatial ecotypes” claims liquid biopsy can predict immunotherapy response. The tumor biology is cool. The statistics are a masterclass in how to do bad biomarker research. @f2harrell has a checklist for this. Let’s go. 1/
1/ Thrilled to share our new paper, out today in @Nature: "Non-invasive profiling of the tumour microenvironment with spatial ecotypes".
Paper (open access): nature.com/articles/s41586-0…
3
11
101
29,636
Apr 29
Regression Modeling Strategies 4-day virtual course May 14-15, 18-19. More than regression:principled flexible modeling, accuracy, interpretation, selection, stability, validation, Bayes, contrast w/ML ... hbiostat.org/course @AmstatNews @InstatsS #Statistics @VUDataScience
9
19
3,852
Apr 28
1-day online course May 11: Intro to R/RStudio & an efficient comprehensive R workflow, rms package, enhancing regression skills to be ready for the RMS 4-day course. Details®istration at hbiostat.org/course @AmstatNews @InstatsS #Statistics #rstats @VUDataScience
5
16
2,764