Causal Inference, Causal Effect Estimation, and Systemic Error Flashcards
Estimating causal effects
understanding the impact of exposure to some factor or treatment on health outcomes
Descriptive epidemiology
communicating the observed world as it is, for example, prevalence of a certain disease in a population
Predictive epidemiology
diagnostic testing (given your characteristics or test results, are you likely to have a particular disease that is not directly observable ). and prognostics. (given your characteristics, are you likely to acquire a particular disease)
What form of epidemiology (descriptive or predictive) is useful for disease surveillance efforts?
Descriptive. It can tell us how prevalent a disease might be, and aid us in the allocation of resources.
What form of epidemiology can help patients understand whether they have a disease, or how their disease will progress over time?
Predictive epi. Typically the realm of clinical epidemiology.
What are the 4 key conditions which together, are sufficient for identification of causal effects in real data?
temporality, causal consistency, exchangeability, no measurement error
Temporality
the condition that things which are causes precede their purported effects in time.
Reverse causality
The (undiagnosed) existence of an outcome affects an individuals exposure status. For example, if early symptoms of COPD lead an individual to quit smoking
Causal consistency
the idea that among people who were exposed, their outcome was no different than it would have been if they had been assigned that exposure (same going for the unexposed). Allows us to move one step from an association to an estimate of causal effect.
What is treatment variation irrelevance and how does it relate to causal consistency?
Consistency is in part the assumption that any variations in the treatment or exposure being studied are irrelevant to the causal mechanism being studied.
Exchangeability (informal)
Informally, it is the condition that study participants who are exposed (or treated) have the same average pre-exposure (or pre-treatment) risk of the outcome as study participants who are unexposed (or untreated).
in other words, if there were no causal effects of the treatment on anyone in the study, would the treated and untreated groups have the same average risk of the outcome?
Exchangeability (formal)
issue of exchangeability is about statistical independence between potential outcomes and the exposure or treatment actually received
What is lack of exchangeability typically due to?
systematic errors, confounding, selection bias
What does “correlation is not causation” typically refer to?
Lack of exchangeability due to systemic errors, confounding, selection bias. Correlation is not not-causation either.
another term for treatment variation irrelevance?
consistency
