Covariate-Adjusted Standardization Method
SAS coding example for implementation of covariate-adjusted standardization method
Reference: O’Brien KM, Upson K, Cook NR and Weinberg CR. “Environmental chemicals in urine and blood: Improving methods for creatinine and lipid adjustment”. In press, Environmental Health Perspectives.
For urinary biomarker analyses where creatinine has been measured at the same time as the biomarker of interest, the covariate-adjusted standardization plus creatinine covariate-adjustment approach performed well. To implement this approach, we first modeled the relationship between log creatinine (logC) and factors known to influence creatinine (e.g. age) using linear regression. This estimates the quantity of creatinine attributable to known factors. The ‘OUTPUT’ statement is used to save the predicted log creatinine values (plogC) in the ‘pred’ file.
OUTPUT out=pred p=plogC;
Because the predicted value accounts for known determinants of creatinine, division of the measured contaminant concentration (E) by the creatinine ratio (observed divided by predicted creatinine) should, theoretically, produce an error-corrected measure of the individual’s exposure level attributable to hydration alone. This is useful because hydration also directly affects the contaminant of interest. Note that we have to exponentiate the predicted log creatinine value to get the predicted creatinine value before calculating the covariate-adjusted exposure value (here E_crt).
MERGE data pred;
BY ID; *files should be sorted by ID;
We can still include creatinine (C) as a covariate in the exposure-outcome regression model to control residual confounding. The adjustment model should include the covariates that affect creatinine in addition to any other confounders. These other confounders are denoted Z1 and Z2. The outcome is disease (D).
MODEL D= E_crt age Z1 Z2 C;
- Clarice R. Weinberg, Acting Branch Chief, Biostatistics & Computational Biology Branch and Principal Investigator
- Tel 919-541-4927