We conceptualized the ERS as a weighted sum of the exposures identified by marginal and conditional analysis, namely, Es and Ec i.e., for subject i, ERSi = ws’Eis + wc’Eic, where ws and wc are vectors of weights corresponding to Es and Ec, respectively. Given that all exposure variables were log-transformed in the present study, the weights (regression coefficients) are on a relative (ratio) scale, not an absolute (difference) scale, and therefore the weights did not need to be scaled. For comparability of the weights on an absolute scale if exposure variables are linearly fit, they need to be scaled (by either standard deviation or IQR).
