Causal inference from meta-epidemiology: a reasonable goal, or wishful thinking?