Get my FREE cheat sheets for Statistics and Research Methods (including transcripts of these lessons) here: www.learnmore365.com/courses/statistics-research-methods-resource-library
While I agree that eliminating confounding, bias, etc. can help demonstrate a causal relationship, I disagree with the dismissal of Hill's criteria. Confounding, bias, and error can really only be controlled in RCTs which are resource intensive and not the most used or common study design. Hill's criteria are broadly applicable to observational studies which make up the bulk of public health and clinical research. While they may not be scientifically robust, they still are pragmatic and relevant.
@samuelrnicholes Bradford hill criteria + excluding bias, chance, confounders, fraud and reverse causation provides a solid systematic approach to infer causality from epidemiology. Some of the Bradford Hill Criteria are essential - temporality and replication in particular. To unequivocally establish causality, you need properly powered, well-designed, well-controlled, well-randomized RCT:s that meet a bunch of criteria. It's very hard to conduct idealistic experiments in reality though.
Could not agree more, and while the Bradford-Hill criteria are not definitive and each criteria has its exceptions/limitations (some more than others), they are a useful framework to discuss causality, especially when there is early, limited or insufficient data, which is often. And don’t think that the fact that confounding exists is sufficient to exclude them. They should be used knowing they are not infallible and that some factors do overlap, but don’t see a better system at present.
there's a lot of problems with this assertion because it allows for low quality data to be used to draw correlations and this is being done all across the health sciences right now. Data quality must be paramount before anything else
This video is basically a word salad to discredit Bradford Hill. That's what people will take from the title. Everything Greg says here is what a responsible data analyst (like Dr. Jessica Rose) would apply prior to publication anyway. Simply logical steps to ensure the reliability of the conclusion. You offer no reason not to use Bradford Hill, only guidelines for it's responsible application. You seem to agree with my assessment in your response to @Efua as well.
Yes change the title The video is great hard to find something beyond basics of correlation vs causation Like OK I get it can someone please get into causation This video was a start
You know how you account for reverse causality loops, without a DeLorean? Bradford #1 Make sure cause precedes effect. Yea, pretty sure that covered it.
Hi Greg, very enjoyable and informative. Question: how would you then go about this process of exclusion (i.e., excluding chance, bias, confounding, reverse causation and fraud). Thanks!
I can think of a million and one examples of things we can never put to RCT. Cigarettes and lung cancer is the easiest example to bring up that even laypeople immediately understand… so if we only trust RCTs, then we lose out on access to a metric shit ton of useful information for our species.
Pretty click-baity title I have to say. All these issues are typically taught along-side Bradford Hill (except Fraud perhaps), so I'm not sure what is being added to the conversation. How does one spot chance or unknown confounders? Well maybe specificity or repetition (with other study types) would help. To be honest I'm not a massive fan of Bradford Hill either.
While this video does present a somewhat valid critique, the author doesn’t do a great job of proposing an alternative system. If you call for tearing down the old structure, I believe it is incumbent upon you to lay out detailed blueprints for how your better structure shall be built.
I guess the Bradford Hill criteria wasn’t good enough for perispinal etanercept injection to be used here in Australia but good enough in the US , any way you work it out
You are great,Mr greg I'm in the last year in medical school and i am interested in global public health , health policy and economics ... But i don't know how to start this pathway ,what and where study degree related to this!! I need your advise !! Thanks
Considering epidemiology is relatively a young science it's not too far fetched to think that 50 years from now Bradford Hill criteria will be replaced by a framework such as what you explained. How would this happen formally, would it be through the publication of a paper and as this is popularly cited and the method engaged more; institutions would move towards use of that criteria, etc?
Interesting point - thanks for the comment. The truth is that Bradford Hill criteria are still useful (just not on their own). Yes I think that publications in the literature will be the first step toward changing thinking.
@@gregmartin hi Greg I see ok thanks. In that instance, are there sequential formulas that are used to ensure the results are robust? I'm trying to understand how the gaps are bridged and before those are formally established what's done in the interim to bring forth reliable results/ conclusion thanks
Get my FREE cheat sheets for Statistics and Research Methods (including transcripts of these lessons) here: www.learnmore365.com/courses/statistics-research-methods-resource-library
Sounds like these guys want to debunk Bradford Hill criteria because the Covid Vaccine injuries very well fits Bradford Hill criteria.
Did they say or imply that
Because they just added more to it they didn't take away anything
While I agree that eliminating confounding, bias, etc. can help demonstrate a causal relationship, I disagree with the dismissal of Hill's criteria. Confounding, bias, and error can really only be controlled in RCTs which are resource intensive and not the most used or common study design. Hill's criteria are broadly applicable to observational studies which make up the bulk of public health and clinical research. While they may not be scientifically robust, they still are pragmatic and relevant.
@samuelrnicholes Bradford hill criteria + excluding bias, chance, confounders, fraud and reverse causation provides a solid systematic approach to infer causality from epidemiology.
Some of the Bradford Hill Criteria are essential - temporality and replication in particular.
To unequivocally establish causality, you need properly powered, well-designed, well-controlled, well-randomized RCT:s that meet a bunch of criteria.
It's very hard to conduct idealistic experiments in reality though.
Could not agree more, and while the Bradford-Hill criteria are not definitive and each criteria has its exceptions/limitations (some more than others), they are a useful framework to discuss causality, especially when there is early, limited or insufficient data, which is often. And don’t think that the fact that confounding exists is sufficient to exclude them.
They should be used knowing they are not infallible and that some factors do overlap, but don’t see a better system at present.
there's a lot of problems with this assertion because it allows for low quality data to be used to draw correlations and this is being done all across the health sciences right now.
Data quality must be paramount before anything else
This video is basically a word salad to discredit Bradford Hill. That's what people will take from the title. Everything Greg says here is what a responsible data analyst (like Dr. Jessica Rose) would apply prior to publication anyway. Simply logical steps to ensure the reliability of the conclusion. You offer no reason not to use Bradford Hill, only guidelines for it's responsible application. You seem to agree with my assessment in your response to @Efua as well.
Yes change the title
The video is great hard to find something beyond basics of correlation vs causation
Like OK I get it can someone please get into causation
This video was a start
Correct, the rooster crowing does not cause the sun to rise. The rising sun, however, does have strong causal relation to the Rooster's crowing.
Thank you
I need more on establishing causation everyone else is just saying and showing what isn't causation
You know how you account for reverse causality loops, without a DeLorean?
Bradford #1 Make sure cause precedes effect.
Yea, pretty sure that covered it.
Hi Greg, very enjoyable and informative. Question: how would you then go about this process of exclusion (i.e., excluding chance, bias, confounding, reverse causation and fraud). Thanks!
Double blinded RCTs.
It's the only way.
Correlation can never imply causation.
I can think of a million and one examples of things we can never put to RCT. Cigarettes and lung cancer is the easiest example to bring up that even laypeople immediately understand… so if we only trust RCTs, then we lose out on access to a metric shit ton of useful information for our species.
Causality loops, nice.
Thank you!
Pretty click-baity title I have to say. All these issues are typically taught along-side Bradford Hill (except Fraud perhaps), so I'm not sure what is being added to the conversation. How does one spot chance or unknown confounders? Well maybe specificity or repetition (with other study types) would help.
To be honest I'm not a massive fan of Bradford Hill either.
While this video does present a somewhat valid critique, the author doesn’t do a great job of proposing an alternative system. If you call for tearing down the old structure, I believe it is incumbent upon you to lay out detailed blueprints for how your better structure shall be built.
It think just adding what he said to do helps a little
And yes we need more am guessing more comes with registration which is understandable
I guess the Bradford Hill criteria wasn’t good enough for perispinal etanercept injection to be used here in Australia but good enough in the US , any way you work it out
You are great,Mr greg
I'm in the last year in medical school and i am interested in global public health , health policy and economics ...
But i don't know how to start this pathway ,what and where study degree related to this!!
I need your advise !!
Thanks
Considering epidemiology is relatively a young science it's not too far fetched to think that 50 years from now Bradford Hill criteria will be replaced by a framework such as what you explained. How would this happen formally, would it be through the publication of a paper and as this is popularly cited and the method engaged more; institutions would move towards use of that criteria, etc?
Interesting point - thanks for the comment. The truth is that Bradford Hill criteria are still useful (just not on their own). Yes I think that publications in the literature will be the first step toward changing thinking.
@@gregmartin hi Greg I see ok thanks. In that instance, are there sequential formulas that are used to ensure the results are robust? I'm trying to understand how the gaps are bridged and before those are formally established what's done in the interim to bring forth reliable results/ conclusion thanks
Your Ice cream and shark attacks metaphor would hard fail Bradford on plausibility, no need to account for confounding on an implausible postulate.
To your last point, fraud; I would say I agree with that more than you'd like.
Good explaination
Thanks
great video!!