Causality. Why you shouldn't use Bradford Hill criteria!

Determining causality isn't easy. Correlation doesn't mean causation. And yet where we see a strong correlation between an exposure and an outcome, we need to be able to determine if there is a cause and effect relationship. Public Health professionals and epidemiologists have typically use the Bradford Hill criteria to establish causality. I don't like the Bradford Hill criteria. Instead I advocate for a process of exclusion (exclude chance, bias, confounding, reverse causation and fraud).

Join this channel to get access to perks:

This channel posts global health and public health teaching videos and videos about how to find the right job in global health. If you haven't already, please consider subscribing to this channel and becoming part of this community.

SUBSCRIBE:
--------------------


LETS CONNECT:
---------------------------
Twitter: @drgregmartin

[samuelrnicholes]

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.

[GayRum]

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.

[loganmiller]

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.

[zrohousing]

Sounds like these guys want to debunk Bradford Hill criteria because the Covid Vaccine injuries very well fits Bradford Hill criteria.

[blurtbum]

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

[patrickelliott]

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!

[bobroberts]

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.

[-med.scenario]

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

[sanadbenali]

Thank you
I need more on establishing causation everyone else is just saying and showing what isn't causation

[kyleflannery]

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.

[bobroberts]

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.

[chasingthesun]

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?

[bobroberts]

Your Ice cream and shark attacks metaphor would hard fail Bradford on plausibility, no need to account for confounding on an implausible postulate.

[bobroberts]

To your last point, fraud; I would say I agree with that more than you'd like.