Single-stage versus two-stage models to estimate creatinine corrected urinary analyte concentrations

The ability to detect 'known' differences in urinary analyte concentrations due to gender, age, and race/ethnicity when adjusted for similar differences in urinary creatinine concentrations were evaluated by a single-stage and a two-stage model by ten simulation studies. Log10 transformed values of observed urinary analyte concentration were used as the dependent variable and age, gender, and race/ethnicity were used as the categorical independent variables. In addition, while single-stage model used log10 transformed values of urinary creatinine as a covariate, two-stage model used a correction factor (CF) determined during the first stage of the model by fitting a secondary model for urinary creatinine. Single-stage model was almost always able to statistically significantly detect 'known' differences due to age, gender, and race/ethnicity. On the other hand, two-stage model was able to statistically significantly detect 'known' differences due to age, gender, and race/ethnicity a maximum of 87.2% of the times and as low as 10.6% of the times primarily because of the presence of multicollinearity between CF and urinary creatinine concentrations. Consequently, as long as the sole objective is to estimate the urinary analyte concentrations adjusted for the effect of all factors including urinary creatinine, single-stage models are the models of choice.