Sunday thoughts:
During our training we physicians learn by rote hundreds to thousands of ultra-rare diagnoses with their associated clinical and biological signs. Comparatively, we spend limited time learning how to interpret clinical trial results and prediction models.

1/15
During our training we physicians learn by rote hundreds to thousands of ultra-rare diagnoses with their associated clinical and biological signs. Comparatively, we spend limited time learning how to interpret clinical trial results and prediction models.

1/15
Most physicians can recite with ease mega-rare differentials with laundry lists of clinical signs, investigations, management.
2/15
2/15
What draws us so strongly to laundry lists? I wonder (I confess I used to be very keen of such lists)
3/15
3/15
The vast majority of these back-of-the-drawer diagnoses we never encounter in our real-life practice. How much energy and effort should we dedicate to diseases most of us will never diagnose or treat?
4/15
4/15
In contrast, many physicians/researchers struggle with the most basic concepts fundamental to interpret biomedical research (eg, p values, immortality bias, classification versus goodness-of-fit metrics, causation versus correlation, conditional probabilities).
5/15
5/15
But those are the bread-and-butter of our practice and research, how we judge the value of the medical literature we read and review, how we inform patients about risks, and most importantly, tools that tell us when to say « I donât know ».
6/15
6/15
I believe most of that struggle can be reduced: physicians need to spend more time learning and « playing » with these concepts. Interactive visualizations can help us achieve this.
7/15
7/15
Some examples:
Type I and II errors: https://shiny.rit.albany.edu/stat/betaprob/
Modeling non-linear effects: https://drjgauthier.shinyapps.io/spliny/
8/15
Type I and II errors: https://shiny.rit.albany.edu/stat/betaprob/
Modeling non-linear effects: https://drjgauthier.shinyapps.io/spliny/
8/15
Statistics are really hard though!Unfortunately a lot harder than laundry lists... Some -many-concepts are still debated. We need our statistician colleagues to help us. It will take a great deal of effort and brain space 
.
9/15


9/15
Yet let us not fret! I believe we can achieve this. Sometimes -often- it will mean going back to the drawing board. Itâs worth it! There has been tremendous advances in the field of data sciences, biostatistics, and machine learning. We need to catch up!
10/15
10/15
IMHO being comfortable with core concepts in biostatistics/biomedical sciences is more important than rote memorization of laundry lists.
11/15
11/15
#statstwitter #epitwitter tweeps I recommend you follow. Iâve learnt so much from them!
@f2harrell @yudapearl @GSCollins @rlmcelreath @daniela_witten @kdpsinghlab @statsepi @_MiguelHernan @PavlosMsaouel @EpiEllie @Lester_Domes @StatModeling @ADAlthousePhD @VickersBiostats
12/15
@f2harrell @yudapearl @GSCollins @rlmcelreath @daniela_witten @kdpsinghlab @statsepi @_MiguelHernan @PavlosMsaouel @EpiEllie @Lester_Domes @StatModeling @ADAlthousePhD @VickersBiostats
12/15
Some great resources:
https://hbiostat.org/bbr/
https://discourse.datamethods.org/
Bayesian statistics:
https://xcelab.net/rm/statistical-rethinking/
Causal inference:
http://bayes.cs.ucla.edu/WHY/
https://www.hsph.harvard.edu/miguel-hernan/causal-inference-book/
13/15
https://hbiostat.org/bbr/
https://discourse.datamethods.org/
Bayesian statistics:
https://xcelab.net/rm/statistical-rethinking/
Causal inference:
http://bayes.cs.ucla.edu/WHY/
https://www.hsph.harvard.edu/miguel-hernan/causal-inference-book/
13/15
Statistical verbiage can sound otherworldly. Hereâs a glossary of statistical terms, courtesy of @f2harrell https://hbiostat.org/doc/glossary.pdf
14/15
14/15
Common statistical myths with references to âpush backâ against old-fashioned reviewers 
https://discourse.datamethods.org/t/reference-collection-to-push-back-against-common-statistical-myths/1787
15/15 [fin]

https://discourse.datamethods.org/t/reference-collection-to-push-back-against-common-statistical-myths/1787
15/15 [fin]