Evaluation of derivative spectroscopy and multivariate calibration for problems in clinical chemistry

Mark Firmer Merrick, Purdue University

Abstract

This thesis describes evaluations of data-processing methods for multiwavelength absorption and fluorescence derivative spectroscopy for multicomponent systems. The multiwavelength derivative spectra were processed by multiple linear regression and the Kalman filter. These approaches to multicomponent determinations are evaluated by using simulated data, measured absorption data for bilirubin and hemoglobin in a serum-like matrix and measured absorption and fluorescence data for porphyrins in diluted urine. In evaluations of these approaches for multicomponent determinations, four specific points addressed are: 1) accuracy and precision as functions of derivative order and width of smoothing window, 2) reduction of low-frequency background signals by using derivative data, 3) modeling of derivative data in the presence of unknown spectral interferents by the adaptive Kalman filter, and 4) reduction of collinearity among spectra by using derivative data. The results for derivative data of one-component systems show that with sufficient smoothing good precision and accuracy can be obtained with multiwavelength and one- and two-wavelength methods. For two-component systems, the multiwavelength method yielded more accurate computed concentrations when compared to one- and two-wavelength methods, and was easier to implement. It was shown that mixtures of bilirubin and hemoglobin and coproporphyrin and uroporphyrin in matrices with low-frequency background signals can be quantified accurately and precisely by using derivative data processed by multiple linear regression. For derivative data processed by the adaptive Kalman filter it was shown that optimum starting wavelengths occur in interferent-free regions of the spectrum; however, identification of these wavelengths in the presence of an unknown interferent was inconsistent when a criterion valid for zeroth-derivative data was used. As an example of an application of the filter, bilirubin was accurately quantified in the presence of unmodeled hemoglobin. Although all five porphyrins can be resolved from simulated data for mixtures by using multiple linear regression with fourth-derivative emission data oscillations in recorded spectra prevented the resolution of the five components. As an alternative, the presence of unmodeled porphyrins was detected and the total amount of porphyrins was determined by processing derivative data by multiple linear regression with models for two and three of the five porphyrins.

Degree

Ph.D.

Advisors

Pardue, Purdue University.

Subject Area

Analytical chemistry

COinS