However, through development of a regression relationship (for model form see Eq. 3), missing concentration data were interpolated using available site-specific concentration and stream discharge measurements.
The missing PM concentrations were imputed because a DLM of PM will be fit to the data, and missing values propagate by up to a factor of 5 when DLMs are used.
For the multivariate analysis only, the missing metabolite concentrations were replaced by a linear combination of the six most correlated analytes according to Kim et al. [48] with the method available in the R package pcaMethods.
The missing PM concentrations were imputed by taking the average of the previous and subsequent day's PM concentration.
Second, the possibility that the cases and the controls differ in the likelihood or in the concentration range of the missing PGE concentrations was examined.
Missing PM2.5 concentrations were imputed using the average of the previous and subsequent days' concentrations.
Missing AAG concentrations were imputed using the Last Observation Carried Forward (LOCF) method.
If > 6 hr (> 25%) of a 24-hr period of PM2.5 concentrations were missing, we set the mean for this 24-hr period of PM2.5 concentration to missing.
Creatinine concentrations were missing for 11 of the first spot urine samples.
