Selectivity is a critical issue for Kromoscopy measurements with broad band filters, particularly for samples composed of multiple solutes with overlapping absorption spectra. The possibility of using Kromoscopy to distinguish similar alcohols, methanol, ethanol and propanol, was investigated over near infrared wavelengths. A set of filters in the first overtone region of the near infrared spectrum was selected and measurements of 50 aqueous mixtures of alcohols were performed to assess the selectivity of Kromoscopy.

Kromoscope results were compared to those using near infrared spectroscopy. Partial least-squares (PLS) regression was used to build calibration models based on near infrared spectra collected with a Fourier transform spectrometer. The prediction errors from the Kromoscopic data were several fold greater than those from the spectroscopic PLS models. A comparison between PLS and net analyte signal regression vectors demonstrates the specificity of the PLS calibration vectors for each alcohol. Selectivity was further documented by applying a pure component selectivity analysis (PCSA).

Monitoring hemodialysis treatments with near infrared spectroscopy is explored with a custom-built acousto optical tunable filter spectrometer. Combination spectra (5000-4000 cm^-1) were collected continuously in spent dialysate during actual dialysis treatments. PLS calibration models for urea were constructed and the robustness of these models was evaluated by comparing prediction performance with different calibration and monitoring sets. Clinically acceptable prediction errors were obtained regardless of the calibration set, which demonstrates the robustness of the PLS-near infrared calibration models.

A calibration model strategy based on the net analyte signal was also used to generate a simple and robust calibration model for monitoring urea in real time. The resulting models provide excellent measurement accuracy over 6 months without the need for complex regression steps.

Two applications of the net analyte signal calibration model for urea were evaluated to estimate of dialysis dose (Kt/V) and to predict hemodialysis-associated complications. Excellent correlation was obtained between Kt/V values predicted from the slope of the urea concentration profile and reference values based on conventional blood measurements. Change in the urea distribution volume suggested that volume drops below a certain level can be used as an indicator of complications occurring during hemodialysis.