Electrochemical signal quantification in saliva: Investigation of signal analysis methods

Saliva has great promise as a background fluid for noninvasive biomarker monitoring at the point of care. However, detection in saliva can be challenging due to its complex composition, which can potentially interfere with analyte signal quantification. In the context of electrochemical sensing, the complex nature of saliva can lead to a high and variable background current such that robust quantification of the analyte signal is challenging. Simple algorithms that work well for quantification in well-defined buffer backgrounds may not be ideal for analysis in complex biofluid backgrounds. To address this, we investigated an analysis method for robust signal extraction from voltammograms measured in unprocessed saliva. Our sequence of voltammogram manipulations consists of (1) log-transformation for improved handling of systematic variation in a highly varying background, (2) smoothing for high-frequency noise reduction, (3) normalization via subtraction of an interpolated piecewise-polynomial spline fit of the analyte-peak-censored voltammogram, and (4) extraction of a peak feature as the analyte signal - peak curvature, peak height, or peak area. In the context of measuring the concentration of the drug carbamazepine in saliva, we systematically determined reasonable parameter values for the manipulations, and evaluated the analysis method using the metrics of the signal coefficient of variation (CV), Welch’s t-statistic, and percent difference between predicted and actual signal. We found that log transformation of the voltammogram current data resulted in overall improved metrics for positive drug concentrations across multiple datasets and for the peak features considered. Comparison across the different peak features indicated that using peak area or peak height resulted in improved resolution and accuracy over peak curvature.