(2 papers I like on this topic: https://t.co/yA4IA5upLO https://t.co/stX8iQkC3x)
@PaulKirvan @je_ff_ff @aca_medic @VPrasadMDMPH Actually, the main result had a P value ~ 0.05, and they did not do an "as treated" analysis, but did an instrumental variable analysis. Here's a paper I wrote on this approach. https://t.co/pvAhspblQJ
@instrumenthull @metrics52 I've been teaching this to doctors for many years. The key reason it hasn't stuck is that in medicine effect sizes just aren't that important. The Q's want to make negative trials positive (nope!), but making small effects a bit
RT @ProfHayward: @kaulcsmc @drjohnm Completely agree that ‘As Treated’ and ‘Per Protocol’ analyses are flawed, but when protocol contaminat…
RT @ProfHayward: @kaulcsmc @drjohnm Completely agree that ‘As Treated’ and ‘Per Protocol’ analyses are flawed, but when protocol contaminat…
@selcukorkmaz One thing I’m never certain about is when to use IV/CACE is a better alternative to per protocol or when is it ever the case that per protocol is better than IV/CACE https://t.co/8UGht93TFC
@PloederlM im surprised they didn't do an IV analysis if they wanted wanted a sense of treatment efficacy without contamination. Much better than doing per-protocol. https://t.co/4ctPdMNe9t
@VPrasadMDMPH @Sensible__Med @drjohnm Altho, I fully agree that the PP analysis is worthless, but an IV analysis would have terrible stat power in this instance & would add nothing. https://t.co/2PVHjL0ovV.
@AleksiReito IV:llä voidaan korjata (tietyillä oletuksilla) RCT-asetelmassa toimenpiteen epätäydellistä noudattamista (non-compliance), joka usein selvästi alle 100%. Se on juuri yksi merkittävä Imbensin ja Angristin kontribuutio, josta hyötyä myös lääket
RT @davidcflood: @raj_mehta @JeremySussman @NoahHaber I have been convinced by @JeremySussman about the role of "contamination-adjusted int…
RT @davidcflood: @raj_mehta @JeremySussman @NoahHaber I have been convinced by @JeremySussman about the role of "contamination-adjusted int…
@raj_mehta @JeremySussman @NoahHaber I have been convinced by @JeremySussman about the role of "contamination-adjusted intention to treat analysis" (IV) to get at patient-level outcomes https://t.co/mVuOIIoTax
@f2harrell @VPrasadMDMPH What about IV estimation? Seems perfect for an RCT with 42% compliance. Deals with the bias of unobserved confounders in the per-protocol model and exclusion restriction seems reasonable here https://t.co/MZpPzmL4gI
@zacksinger26 @djc795 @JeremySussman @mikejohansenmd Adding the paper here for reference. https://t.co/EA8FUCeFGG
@prof_goldberg @willyhfrick @eemoin That's a good summary of the trade-off, yes. My feeling is you've invested a lot in the RCT bc you don't trust observational data, so you should use it. In the grand spirit of self-promotion, there are workarounds, thoug
@ADiederichsen @kaulcsmc @mmamas1973 @pash22 @djc795 @MadalinaGarbi @dr_benoy_n_shah @VictorDayan1 @Argenscore @ovidiogarciav @SABOURETCardio @LRussellWolpe @DavidLBrownMD @NHSHealthChecks @dompagano @drjohnm Seems this would be a great application of a CA
@ADAlthousePhD @ashtroid22 @avicenna 4/ IMO, quant adjusting to get ATE from ITT is not always feasible, but an IV approach is sometimes feasible. Some VERY GOOD stats/epi ppl believe in sophisticated PP analyses to get valid ATE, but I don't (perhaps I'm
@johngraves9 I co-wrote this to encourage it. It’s a harder lift than I’d expected. https://t.co/r5cyOsv0UA
RT @JeremySussman: @statsepi Whenever I talk about this paper, they want IV to make their null study positive and lose interest when I say…
RT @JeremySussman: @statsepi Whenever I talk about this paper, they want IV to make their null study positive and lose interest when I say…
RT @JeremySussman: @statsepi Whenever I talk about this paper, they want IV to make their null study positive and lose interest when I say…
@statsepi Whenever I talk about this paper, they want IV to make their null study positive and lose interest when I say that mostly it’ll make their positive study have a larger effect size. https://t.co/nwVN8HJ8mS
I ought to be sleeping Instead I'm up trying to decide if @ProfHayward + @JeremySussman "using instrumental variables to adjust for treatment contamination in RCTs" (https://t.co/28oqfuKPki) can be generalized to target trials with post-"randomization" cr
@ajaykirtane @DavidLBrownMD @michaelTCTMD @kaulcsmc This feels like a moment when @JeremySussman would normally pop in to suggest the utility of CA-ITT analyses to improve the estimate of the treatment effect of revascularization and cite this great paper
We tend to apply RCTs results broadly, but they only apply to pts who followed trial protocol. Pretty straightforward, right? Well, our IV results apply only to STEMI pts who received ICU care only because they lived close to a high ICU hospital. https://
@CardiacJoshi @DrMarkMills @bogdienache @DrAfzalSohaib There's an alternative to PP. You can estimate the Avg Treatment Effect (ATE) using instrumental variable analysis. 👇 https://t.co/g4DtiCPZ9o
@YFeyman 🤷♂️ Article below is almost 10-yrs now 👇 https://t.co/AQLau6c90O
@YFeyman 8 years of discussions about this. Big picture: IV increases effect size but doesn’t change p values much. But effect size rarely matter in medicine - any proven benefit is usually enough. https://t.co/i4LQtaQEXO
RT @ProfHayward: @amitabhchandra2 It's becoming less common, but AT & PP analysis are still done. No disagreement at all w/ ur assessment,…
@amitabhchandra2 It's becoming less common, but AT & PP analysis are still done. No disagreement at all w/ ur assessment, tho. 👇👇 https://t.co/AQLau6c90O
@amitabhchandra2 @haroldpollack Nice discussion of this from @ProfHayward @JeremySussman https://t.co/lbDdwf2eC5
Might need more nuance in case native/SVG unamenable to #PCI. Maybe have to re-randomize BMS vs DES once randomized to SVG arm. Then which DES? Would need some input from @rwyeh on treatment contamination https://t.co/FJkWc96vQy
https://t.co/8UGht93TFC Contamination Adjusted Intent to Treat .... or using treatment assignment as an instrument #econometrics #causalinference #biostatistics #healthcare #economics
RT @ProfHayward: @_MiguelHernan @AlexanderZheut1 2. There is no place in a civilized world for `per protocol' analysis! Instrumental variab…
RT @ProfHayward: @_MiguelHernan @AlexanderZheut1 2. There is no place in a civilized world for `per protocol' analysis! Instrumental variab…
@_MiguelHernan @AlexanderZheut1 2. There is no place in a civilized world for `per protocol' analysis! Instrumental variable analysis should be used to adjust the ITT result for Rx contamination (usually called the Avg Treatment Effect [ATE]). https://t.c
@WalkeyAllan @davidlederer @iwashyna @DrSamuelBrown @_MiguelHernan The paper from @_MiguelHernan in NEJM is amazing. Regarding IV when analyzing RCTs, I also like this from BMJ https://t.co/ooLtHlcYhU
@davidlederer @IPFdoc @biorxivpreprint I defer these questions to @JeremySussman, esp when I am as jet lagged as I am this morning. His @bmj_latest paper still rocked my world about this: https://t.co/r3isqFYqJO
@gsmartinmd Might there be a role for Instrumental Variable Analysis to adjust for treatment cross contamination? https://t.co/Waa9yM2chl
RT @ProfHayward: @MGtberg @rallamee @mshunshin @ProfDFrancis @RogueRad @venkmurthy @ajaykirtane @rwyeh The problem is well known, & appropr…
@MGtberg @rallamee @mshunshin @ProfDFrancis @RogueRad @venkmurthy @ajaykirtane @rwyeh The problem is well known, & appropriate, & inappropriate, methods are well described. per protocol, etc tends to biases results towards a positive effect, &
really nice review of instrumental variable analysis and its role vis a vis RCTs by @JeremySussman & @ProfHayward. they’re not perfect, but it’s hard to understand why per-protocol analyses are even reported anymore when these techniques exist https:/
RT @ProfHayward: @MariaGlymour @ProfDFrancis Absolutely! Per Protocol and As Treated analysis have no place in a civilized world! 👇🏾👇🏾👇🏾👇🏾…
RT @venkmurthy: To all those discussing as treated analysis from #CABANA, DON'T. Instrumental variables isn't all that complicated to do a…
RT @ProfHayward: 2 takes on #CABANA: 1. It is very under-powered. HR of 0.85 for this composite would be huge, IF TRUE. Also, cross-over b…
RT @ProfHayward: 2 takes on #CABANA: 1. It is very under-powered. HR of 0.85 for this composite would be huge, IF TRUE. Also, cross-over b…
RT @ProfHayward: @drjohnm @leftbundle @f2harrell @ProfDFrancis I hate 2B harsh, but IMO OTs inherent flaws are well enough documented that…
RT @ProfHayward: @drjohnm @leftbundle @f2harrell @ProfDFrancis I hate 2B harsh, but IMO OTs inherent flaws are well enough documented that…
@drjohnm @leftbundle @f2harrell @ProfDFrancis I hate 2B harsh, but IMO OTs inherent flaws are well enough documented that continuing to conduct OT analysis is scientific negligence. I'm sorry, but there comes a point when ignoring Semmelweis & failure
RT @ProfHayward: 2 takes on #CABANA: 1. It is very under-powered. HR of 0.85 for this composite would be huge, IF TRUE. Also, cross-over b…
RT @ProfHayward: 2 takes on #CABANA: 1. It is very under-powered. HR of 0.85 for this composite would be huge, IF TRUE. Also, cross-over b…
@DanielAlyeshmer @ProfHayward My guess is that @ProfHayward supports it. https://t.co/qfiM1k5WTY
@ADAlthousePhD @rwyeh Start there and then go read @JeremySussman and @ProfHayward brief paper explaining it. https://t.co/dv8pOzM3pw
RT @ProfHayward: @MariaGlymour @ProfDFrancis Absolutely! Per Protocol and As Treated analysis have no place in a civilized world! 👇🏾👇🏾👇🏾👇🏾…
RT @venkmurthy: To all those discussing as treated analysis from #CABANA, DON'T. Instrumental variables isn't all that complicated to do a…
RT @ProfHayward: 2 takes on #CABANA: 1. It is very under-powered. HR of 0.85 for this composite would be huge, IF TRUE. Also, cross-over b…
RT @venkmurthy: To all those discussing as treated analysis from #CABANA, DON'T. Instrumental variables isn't all that complicated to do a…
RT @venkmurthy: To all those discussing as treated analysis from #CABANA, DON'T. Instrumental variables isn't all that complicated to do a…
To all those discussing as treated analysis from #CABANA, DON'T. Instrumental variables isn't all that complicated to do and is just a much better way of dealing with crossovers. @drjohnm @JeremySussman @ProfDFrancis @ProfHayward @bnallamo https://t.co/m
RT @ProfHayward: 2 takes on #CABANA: 1. It is very under-powered. HR of 0.85 for this composite would be huge, IF TRUE. Also, cross-over b…
RT @ProfHayward: 2 takes on #CABANA: 1. It is very under-powered. HR of 0.85 for this composite would be huge, IF TRUE. Also, cross-over b…
RT @ProfHayward: 2 takes on #CABANA: 1. It is very under-powered. HR of 0.85 for this composite would be huge, IF TRUE. Also, cross-over b…
RT @ProfHayward: 2 takes on #CABANA: 1. It is very under-powered. HR of 0.85 for this composite would be huge, IF TRUE. Also, cross-over b…
RT @ProfHayward: 2 takes on #CABANA: 1. It is very under-powered. HR of 0.85 for this composite would be huge, IF TRUE. Also, cross-over b…
2 takes on #CABANA: 1. It is very under-powered. HR of 0.85 for this composite would be huge, IF TRUE. Also, cross-over biases the point estimate towards the null AND hurts power further. 2. IGNORE the per protocol analysis = scientific malfeasance 😡 htt
RT @JeremySussman: @ProfDFrancis @rwyeh @FelixValencia10 @willsuh76 @drjohnm To me, they estimate different things. ITT - effect of randomi…
@WFMMD @atulverma_md One option is to try to correct using a “contamination adjusted analysis”. It’s potentially a good middle ground for trials with substantial cross over. https://t.co/TiuZ0x3HWj
@MariaGlymour @ProfDFrancis Absolutely! Per Protocol and As Treated analysis have no place in a civilized world! 👇🏾👇🏾👇🏾👇🏾👇🏾 https://t.co/D7n2zsmvXJ
@ProfDFrancis There are ways to handle this that are effective. As @JeremySussman @ProfHayward and others have written, instrumental variables approach can address this. https://t.co/hrLtXBnm3h
RT @JeremySussman: At the sort-of-request of @ProfDFrancis, some tweets on analyzing RCTs with instrumental variables. For those who like p…
RT @JeremySussman: At the sort-of-request of @ProfDFrancis, some tweets on analyzing RCTs with instrumental variables. For those who like p…
RT @ProfDFrancis: Excellent description of instrumental variable analysis by Jeremy Sussman. I didn't really know what it was before readin…
RT @ProfDFrancis: Excellent description of instrumental variable analysis by Jeremy Sussman. I didn't really know what it was before readin…
Very nice tweetorial: demonstrates that per-protocol/as-treated inflate effect, and ITT gives a conservative estimate of it https://t.co/gG8xs0J4KQ
RT @JeremySussman: At the sort-of-request of @ProfDFrancis, some tweets on analyzing RCTs with instrumental variables. For those who like p…
Excellent description of instrumental variable analysis by Jeremy Sussman. I didn't really know what it was before reading his thread and paper. (Of course that didn't stop me pontificating.) https://t.co/dFISaVHvdH
RT @ProfHayward: Nice thread by @JeremySussman on how to deal with non-adherence and cross-over in RCTs 👇🏾 https://t.co/nn1Pem4YBu
RT @ProfHayward: Nice thread by @JeremySussman on how to deal with non-adherence and cross-over in RCTs 👇🏾 https://t.co/nn1Pem4YBu
RT @ProfHayward: Nice thread by @JeremySussman on how to deal with non-adherence and cross-over in RCTs 👇🏾 https://t.co/nn1Pem4YBu
Nice thread by @JeremySussman on how to deal with non-adherence and cross-over in RCTs 👇🏾 https://t.co/nn1Pem4YBu
RT @JeremySussman: At the sort-of-request of @ProfDFrancis, some tweets on analyzing RCTs with instrumental variables. For those who like p…
RT @JeremySussman: At the sort-of-request of @ProfDFrancis, some tweets on analyzing RCTs with instrumental variables. For those who like p…
RT @JeremySussman: At the sort-of-request of @ProfDFrancis, some tweets on analyzing RCTs with instrumental variables. For those who like p…
RT @JeremySussman: At the sort-of-request of @ProfDFrancis, some tweets on analyzing RCTs with instrumental variables. For those who like p…
RT @JeremySussman: At the sort-of-request of @ProfDFrancis, some tweets on analyzing RCTs with instrumental variables. For those who like p…
At the sort-of-request of @ProfDFrancis, some tweets on analyzing RCTs with instrumental variables. For those who like paragraphs, see my old paper: https://t.co/2muzQbBXOA https://t.co/u8pVtida1X
... intention to treat analysis answers the question: “how much do study participants benefit from being assigned to a treatment group?” https://t.co/6BRNInroYy
RT @JeremySussman: @ProfDFrancis @rwyeh @FelixValencia10 @willsuh76 @drjohnm To me, they estimate different things. ITT - effect of randomi…
@rwyeh @JeremySussman @FelixValencia10 @willsuh76 @drjohnm Sorry not at ACC. But I remembered Jeremey Sussman wrote a paper on it so I will read rather than randomly spouting on from a position of ignorance. Spent too long speaking to statin denialists! h