Urine Glucose Levels

Return to Smart Healthcare

Using Digital Filters to Obtain Accurate
Trended Urine Glucose Levels from
Toilet-Deployable Near-Infrared Spectrometers

Many over-the-counter glucose measurement systems currently exist but are not widely used by nondiabetic consumers because of the inconvenience. There exists a need for new methods of conveniently detecting early stages of diabetic or prediabetic conditions rather than waiting for the disease to progress to the point that symptoms indicative of physiological damage are present and a user requests medical care. Near-infrared (NIR) spectroscopic urinalysis has shown some promise for use as an unobtrusive measurement system for glucose levels but has required expensive equipment. This paper presents a method of combining a cost-effective, home-deployable NIR system with a non-traditional trend-based data analysis to extract representative glucose levels from patients. By taking multiple measurements over time with an unobtrusive, automatic, in-toilet urinalysis system, limited accuracy samples from each patient can be averaged to obtain an improved accuracy trended value. Data trending is able to predict glucose levels with sufficient accuracy to be clinically relevant in the detection of chronically high glucose conditions. The bandwidth, or averaging window, of the filters can be varied to achieve a target accuracy level, even when the error of individual measurements is large and variable. Urine spectra can be captured from an at-home or at-work toilet with a urine capture slot and NIR spectrometer. A new data reporting strategy is proposed for trended measurements, whereby filtered data is reported with a known and acceptable post-filter variance, rather than reporting individual sample measurements. This is in contrast to traditional methods of single-point clinical tests, which may require expensive equipment to achieve sufficient single-point accuracy, be obtrusive or inconvenient, available only on demand, or susceptible to outliers.


NIR spectroscopy promises in-toilet urinalysis, providing more ubiquitous and unobtrusively obtained information than semi-quantitative colorimetric assay urine tests. While the compact NIR instrumentation used for this study lacks the sensitivity and stability to detect normal levels of urine glucose with a single measurement, by averaging data from sequential urine samples, an accurate glucose level trend can be obtained. This scheme for averaging many samples to achieve an improved SNR is a completely different approach from traditional diagnoses that rely on presumed accurate single measurements. Significantly, the variance of the filtered data can be monitored and the filter bandwidth adjusted such that the trended result meets a desired level of accuracy even when the individual sample measurements cannot. This departure from single-point measurement reporting to trended data reporting enables remote preventative care and unobtrusive patient monitoring and is especially useful for health trends which change slowly. Notably, the filter performance can be optimized to provide sufficient averaging to achieve a target accuracy level or optimized to follow sharp trends in the data. The trended measurements of multiple samples over time may provide a better overall picture of a user’s medical progress than isolated lab samples that are not robust against outlier data. Medical diagnostics using trended data with validated and tunable accuracy can expand the role of health tracking and disease management to a new level of usefulness and cost efficacy. This paper has focused on trended glucose levels but the concept of data trending over sequential measurements could also be applied to a variety of medical pre-screening. Several other analytes have been measured by the authors with the same spectrometer setup described in this paper and may be discussed in future papers. 

View the PDF Abstract >