Assessing the validity of principal component regression models in different analytical conditions

This study proposes a methodology for assessing the validity of principal component regression models when the experimental conditions which have been used in the process of modeling may have changed. The methodology proposed is based on the procedure for selecting the validation sample subset which includes the D-optimal criterion and application of Fedorov's exchange algorithm. Two basic performance characteristics define the validity of the models: trueness is assessed by linear regression using the joint confidence test for the slope, and the intercept and precision is estimated by bias corrected MSEP and RRMSEP. The methodology is validated with a simulated data set and three real data sets corresponding to models constructed for spectrophotometric data from determinations of various analytes in waters using sequential injection analysis (SIA). Using a reduced number of samples can be very useful in several applications, such as in process analytical control, and is especially useful as an initial step to check the need for standardization.