Multivariable regression models typically assume a causal structure such as A, whereas a more realistic causal structure is depicted in B. X, exposure; Y, outcome; C, confounders; M, mediators; Z, colliders.