Roger Borràs
@RWJE_BA
Followers
976
Following
9K
Media
262
Statuses
6K
Mathematician, biostatistician, gamer, skier, violinist & R lover. Evidence-Based Medicine. Hospital Clínic de BCN. @idibaps @CIBER_ESP. Professor @UABBarcelona
Terrassa
Joined November 2016
Truth starts with accurate words. 'Emulate' means to copy. Target trial emulation does not copy RCT design other than defining the inception cohort. It does not copy randomization, blinding, prospective data collection, specifying adjustment variables before collecting data ..
“Emulating” RCTs is disastrous language applied to an infeasible situation - how does one emulate blinding? randomization? The old observational data is dressed up as the new “emulated RCT” Marketing terminology and salesmanship is all this is imho @RWJE_BA
5
25
90
@dylanarmbruste3 @pash22 @garybassmd @ajlees @adamcifu @RWJE_BA In my subjective personal opinion that paper has not aged exceptionally well. I don’t see propensity scores as doing causal inference per se, instead just providing a filter to reduce the dimensionality of all MEASURED confounders to one.
3
4
4
@garybassmd @ajlees @adamcifu @f2harrell ... And Roger @RWJE_BA adds that it does not copy placebo effects, time and space concurrent controls, credible pre-specification, ITT/PP comparison, control for regression to the mean, mathematical known satistical laws for inference... Randomization is "emulated" with PS in
2
3
6
@yudapearl I never said that it’s necessary for trialists to follow a philosophy of CI. They just need to know how to design controlled experiments that rule out all explanations save the experimental manipulation. No symbolism needed before secondary questions requiring counterfactuals.
2
2
12
👀🔥 BREAKING ‼️ Coge fuerza las informaciones Next Gen de Xbox que llevamos trayendo al canal durante más de 1 año y medio. Windows Central corrobora que: será un PC con interfaz de consola, podrás usar Steam y otras tiendas, tendrás multiplayer gratis (en teoría) y
40
57
870
@JonNichEdwards @5_utr @yudapearl @PatientStormDoc @RWJE_BA @doc_BLocke @stephensenn This was addressed earlier. The target pop will have a different covariate distribution and we don’t care unless there are A-B differences in unmodeled interacting (with treatment) factors. That’s why a key estimand for binary Y is P(Y=1|X, B) - P(Y=1|X, A) which is CTE.
2
2
3
Are #RelativeRisks overwhelmingly similar across risk groups? And if so, can we personalize care by applying relative risk to individual patients’ risk profiles? This key #EBM study demonstrates that the answer to both questions is an unquestionable YES. https://t.co/geRsjPPZab
pubmed.ncbi.nlm.nih.gov
The fixed effects OR, random effects OR and random effects RR appear to be reasonably constant across different baseline risks. Given the interpretational and arithmetic ease of RR, clinicians may...
1
9
29
@f2harrell @mzloteanu @PatientStormDoc @Stefan11005620 @RWJE_BA @yudapearl @doc_BLocke @5_utr Yes! Almost 90% of the Obs studies are NOT designed for the question they are asking...they are simply using a "pre-existing data" which was designed for other questions.
2
1
3
@mzloteanu @PatientStormDoc @Stefan11005620 @RWJE_BA @yudapearl @doc_BLocke @5_utr I plan to write a very opinionated blog article about this sometime. Causal inference comes from design, bias-free measurements, and effective sample size. Non-designed observational studies are almost always futile for estimating intervention effects.
4
5
15
@gv_lazcano @AdanZBecerra1 @kat_hoffman_ @RWJE_BA @dylanarmbruste3 @KordingLab @ildiazm @yudapearl @elmir1omerovic @dredgardorsey @ArgoloF @drjohnm @kaulcsmc @GreggWStone @_MiguelHernan You are exactly correct. Randomized trials have been done using sham brain surgery, sham heart surgery, sham knee surgery, and sham bronchial catheter ablation. If you can randomize in those contexts you can randomize almost anywhere. #clinicaltrial #biostatistics
2
8
41
@raj_mehta @RWJE_BA @f2harrell @doc_BLocke @PatientStormDoc Misplaced concern about RCTs not generalization >>> RCTs not generalizing Patients aren’t forced on RCTs nor are there sampling weights etc - the idea of generalizing to population (PATE) is not the point of RCTs
2
1
5
-El futuro de PlayStation es bastante oscuro. •PS6 no destacará en potencia •Su enfoque será en usuarios activos •Con Xbox Magnus + Steam, PS6 se vuelve innecesaria •Menos exclusivos. -Xbox dará el salto y pinta brillante; una pena lo de los pitufos, pero así es esto xd
187
29
274
9
216
4K
Ordinal semiparametric cumulative probability models have an amazing variety of statistical methods as special cases, including proportional odds and proportional hazards models - https://t.co/ZH0TkafIUw
#Statistics
0
16
78
@doc_BLocke @PatientStormDoc @RWJE_BA @5_utr You are right. I’ve never found an interaction by exploratory analysis (not driven by any subject matter knowledge) that replicated. And RCTs are barely powered to detect even an overall treatment effect.
2
1
3
Immediate separation with parallel trends thereafter means this utter BS Waste of so much effort How does this get funded & published in the BMJ?
SGLT-2 inhibitors were associated with an 11% lower risk of autoimmune rheumatic diseases compared with sulfonylureas, finds new study https://t.co/ao4Juw1Oub
12
16
120
Porfin tienes un rato libre para jugar... La actualización 😂
75
60
2K