To infer causality you need interventions. Although there are some methods to infer causal relationships from observational data (e.g. see Pearl or Spirtes' wor

To infer causality you need interventions. Although there are some methods to infer causal relationships from observational data (e.g. see Pearl or Spirtes' work, which is ignored by most epidemiologists). I think the biggest problem is that they are not measuring low level molecular variables like metabolites, or abundances of gut bacteria. Hence, it's pretty hard to infer a causal chain. And since controlling for covariates is so hard, inferring something useful is difficult.