Interfering substances for glucose sensing by near-infrared spectroscopy

Near-Infrared (NIR) Spectroscopy could be an alternative method to measure blood glucose concentration, with access to low-cost components and potential for non-invasive use. Possible interferents to this method are not widely investigated. We aimed to identify lactate, ethanol, acetaminophen and caffeine for possible interference. Glucose and the four interferents were mixed in varying physiological concentrations and measured by NIR spectroscopy in vitro. A calibration model was built by partial least squares regression. The full set (273 spectra) was divided into subsets (59-73 spectra) where one of the interferents were left out and new models were built without knowledge of the interferent. Subsets with random samples were used as a reference to the smaller calibration sets of the interferents. All models were applied to a validation set containing all interferents. The models built without ethanol and caffeine had high errors and several points outside of zone A and B in the Clarke Error grid. The model without lactate mistook lactate for glucose and gave false high predictions. The model without ethanol gave lower glucose predictions with increasing ethanol concentration. The models without acetaminophen and caffeine were comparable to the random subsets. Lactate and ethanol interfere with glucose prediction by NIR spectroscopy if not included in the model calibration. Acetaminophen and caffeine are not interferents to glucose using NIR spectroscopy.