Thread by @teddyhla

NEXT : @AriErcole
Association is not necessarily a causation
RCTs are thought of as "gold standard" for a good reason.
"Randomisation" eliminates influences of confounders.
- allows "causality" inference.

@AriErcole RCTs require relatively little prior knowledge.
But
"DO WE HAVE ASSURANCE OF THIS" ?
we try to "by having a inclusion criteria"
🧐 we can only control what we know

RCTs have limitations - dont really imply causality absolutely.

@ESICM #datascience #ai #ml #icudata #RCT

🗒 Structural Causal Models
- multiple regression models
can make these DAGs arbitrarily complex (many regressions = structural equation modelling)
see pic

Example of DAG

📃 What about "Natural Experiments"
- in a big enough data set, there will be a "digital twin"
- creates a natural experiment if we hav enough data
- can use to address confounding by indication
e.g., Propensity score matching(PSM) and Inverse probability of Treatment Weighting

What do those mean - and here

📃How about "time-series" causality?
- given t-s e.g., HR
- all physiology fluctuates
- causality based on "information flow between signals as it has "DIRECTIONALITY"
-eg fluctuation has meaning
- signal complexity outperforms anything either as early predictors of sick or not.

Example of this.

📃What about "reinforcement learning"?
give computer "policies" and see what is the "best policy". a bit like "propensity score matching"
- ML model
mentions @matkomorowski paper

What about “individualised treatment effects” using @MihaelaVDS clairvoyance models

@MihaelaVDS Summary
- statistical techniques nor RCTs prove causality
- "natural experiments" are another way of assessing causality
- "big data" allows sophisticated processes #datascience #ai #ml #icudata #LIVES2022