METHOD OF SAMPLE CONTROL AND CALIBRATION ADJUSTMENT FOR USE
WITH A NONINVASIVE ANALYZER
BACKGROUND OF THE INVENTION
FIELD OF THE INVENTION
The invention relates to noninvasive determination of analytes in the human body.
More particularly, the invention relates to methods of anaiyte calibration and
prediction that adjust for state changes not compensated for with a static calibration
model.
DESCRIPTION OF RELATED ART
NONINVASIVE
A spectroscopy based noninvasive analyzer delivers external energy in the form of
light to a region of the body where the photons interact with the chemical
constituents and physiology of the sampled tissue. A portion of the incident photons
are scattered or transmitted out of the body where they are detected. Based upon
knowledge of the incident photons and detected photons, the chemical and/or
structural basis of the sampled site is elucidated. A distinct advantages of a
noninvasive system includes the determination of a chemical constituent
concentration in the body without the generation of a biohazard in a pain free
manner and with the use of limited consumables. Further the technique allows for
multiple anaiyte concentrations to be determined at one time. Some common
examples of noninvasive analyzers are magnetic resonance imaging (MRI), X-rays,
pulse oximeters, and noninvasive glucose analyzers. With the exception of X-rays,
these determinations are performed with relatively harmless wavelengths of
radiation. Examples herein focus on noninvasive glucose concentration
determination, but the principles apply to the detection of other analytes, such as
fats, proteins, water, and blood or tissue constituents.
DIABETES
Diabetes is a chronic disease that results in improper production and utilization of
insulin, a hormone that facilitates glucose uptake into cells. While a precise cause of
diabetes is unknown, genetic factors, environmental factors, and obesity appear to
play roles. Diabetics have increased risk in three broad categories: cardiovascular
heart disease, retinopathy, and neuropathy. Diabetics often have one or more of the
following complications: heart disease and stroke, high blood pressure, kidney
disease, neuropathy (nerve disease and amputations), retinopathy, diabetic
ketoacidosis, skin conditions, gum disease, impotence, and fetal complications.
Diabetes is a leading cause of death and disability worldwide. Moreover, diabetes is
merely one among a group of disorders of glucose metabolism that also include
impaired glucose tolerance, and hyperinsulinemia, or hypoglycemia.
DIABETES PREVALENCE AND TRENDS
Diabetes is an ever more common disease. The World Health Organization (WHO)
estimates that diabetes currently afflicts 154 million people worldwide. There are 54
million people with diabetes living in developed countries. The WHO estimates that
the number of people with diabetes will grow to 300 million by the year 2025. In the
United States, 15.7 million people or 5.9 per cent of the population are estimated to
have diabetes. Within the United States, the prevalence of adults diagnosed with
diabetes increased by six percent in 1999 and rose by 33 percent between 1990 and
1998. This corresponds to approximately eight hundred thousand new cases every
year in America. The estimated total cost to the United States economy alone
exceeds $90 billion per year. Diabetes Statistics, National Institutes of Health,
Publication No. 98-3926, Bethesda, MD (November 1997).
Long-term clinical studies show that the onset of diabetes related complications is
significantly reduced through proper control of blood glucose concentrations. The
Diabetes Control and Complications Trial Research Group, The effect of intensive
treatment of diabetes on the development and progression of long-term
complications in insulin-dependent diabetes mellitus, N Eng J of Med , 329:977-86
(1993); U.K. Prospective Diabetes Study (UKPDS) Group, Intensive blood-glucose
control with sulphonylureas or insulin compared with conventional treatment and risk
of complications in patients with type 2 diabetes, Lancet, 352:837-853 (1998); and
Y. Ohkubo, H. Kishikawa, E. Araki, T. Miyata, S. Isami, S. Motoyoshi, Y. Kojima, N.
Furuyoshi, M. Shichizi, Intensive insulin therapy prevents the progression of diabetic
microvascular complications in Japanese patients with non-insulin-dependent
diabetes mellitus: a randomized prospective 6-year study, Diabetes Res Clin Pract,
28:103-117 (1995).
A vital element of diabetes management is the self-monitoring of blood glucose
concentrations by diabetics in the home environment. However, current monitoring
techniques discourage regular use due to the inconvenient and painful nature of
drawing blood through the skin prior to analysis. The Diabetes Control and
Complication Trial Research Group, supra. As a result, noninvasive measurement of
glucose concentration is identified as a beneficial development for the management
of diabetes. Implantable glucose concentration analyzers coupled to an insulin
delivery system providing an artificial pancreas are also being pursued.
SAMPLING METHODOLOGY
A wide range of technologies serve to analyze the chemical make-up of the body.
These techniques are broadly categorized into two groups, invasive and
noninvasive. For the purposes of this document, a technology that acquires any
biosample from the body for analysis or if any part of the measuring apparatus
penetrates into the body, the technology is referred to as invasive.
• Invasive: Some examples of invasive technologies for glucose concentration
determination in the body are those that analyze the biosamples of whole
blood, serum, plasma, interstitial fluid, and mixtures or selectively sampled
components of the aforementioned. Typically, these samples are analyzed
with electrochemical, electroenzymatic, and/or colorimetric approaches. For
example, enzymatic and colorimetric approaches are used to determine the
glucose concentration in interstitial fluid samples.
• Noninvasive: A number of approaches for determining the glucose
concentration in biosamples have been developed that utilize
spectrophotometric technologies. These techniques include: Raman and
fluorescence, as well as techniques using light from the ultraviolet through the
infrared [ultraviolet (200 to 400 nm), visible (400 to 700 nm), near-IR (700 to
2500 nm or 14,286 to 4000 cm"1), and infrared (2500 to 14,285 nm or 4000 to
700 cm-1)].
NONINVASIVE GLUCOSE DETERMINATION
There exist a number of noninvasive approaches for glucose concentration
determination. These approaches vary widely, but have at least two common steps.
First, an apparatus is used to acquire a signal from the body without obtaining a
biological sample. Second, an algorithm is used to convert this signal into a glucose
concentration determination.
One type of noninvasive glucose concentration analyzer is based upon spectra.
Typically, a noninvasive apparatus uses some form of spectroscopy to acquire a
signal or spectrum of a body part. Utilized spectroscopic techniques include Raman
and fluorescence, as well as techniques using light from the ultraviolet through the
infrared [ultraviolet (200 to 400 nm), visible (400 to 700 nm), near-IR (700 to 2500
nm or 14,286 to 4000 crrr1), and infrared (2500 to 14,285 nm or 4000 to 700 cm-1)].
A particular range for noninvasive glucose determination in diffuse reflectance mode
is about 1100 to 2500 nm or ranges therein. K. Hazen, Glucose Determination in
Biological Matrices Using Near-Infrared Spectroscopy. doctoral dissertation,
University of Iowa (1995).
Mode
Typically, three modes are used to collect noninvasive spectra: transmittance,
transflectance, and/or diffuse reflectance. For example the signal collected, typically
being light or a spectrum, is transmitted through a region of the body such as a
fingertip, diffusely reflected, or transflected. Transflected here refers to collection of
the signal not at the incident point or area (diffuse reflectance), and not at the
opposite side of the sample (transmittance), but rather at some point on the body
between the transmitted and diffuse reflectance collection areas. For example,
transflected light enters the fingertip or forearm in one region and exits in another
region typically 0.2 to 5 mm or more away depending on the wavelength used.
Thus, light that is strongly absorbed by the body such as light near water absorbance
maxima at 1450 or 1950 nm is collected after a small radial divergence and light that
is less absorbed such as light near water absorbance minima at 1300, 1600, or 2250
nm is collected at greater radial or transflected distances from the incident photons.
Site
Noninvasive techniques are not limited to using the fingertip as a measurement site.
Alternative sites for taking noninvasive measurements include: a hand, finger,
palmar region, base of thumb, wrist, dorsal aspect of the wrist, forearm, volar aspect
of the forearm, dorsal aspect of the forearm, upper arm, head, earlobe, eye, tongue,
chest, torso, abdominal region, thigh, calf, foot, plantar region, and toe.
Instrumentation
While this specification focuses on optical based noninvasive analyzers, it is
important to note that noninvasive techniques do not have to be based upon
spectroscopy. For example, a bioimpedence meter is considered to be a
noninvasive device. Within the context of the invention, any device that reads a
signal from the body without penetrating the skin and collecting a biological sample
is referred to as a noninvasive glucose analyzer. For example, a bioimpedence
meter is a noninvasive device.
Noninvasive glucose concentration determination using a near-infrared analyzer
generally involves the illumination with an input element of a small region on the
body with near-infrared (NIR) electromagnetic radiation, infrared light in the
wavelength range 700 to 2500 nm, or one or more wavelength ranges therein, such
as 1100 to 1800 nm. The light is partially absorbed and partially scattered according
to its interaction with the constituents of the tissue prior to being reflected back to
light collection means optically coupled to a detector or directly to a detector. The
detected light contains quantitative information that corresponds to the known
interaction of the incident light with components of the body tissue including water,
fat, protein, and glucose.
A noninvasive glucose concentration analyzer has one or more beam paths from a
source to a detector. Light source types include a blackbody source, a tungsten-
halogen source, one or more LED's, and one or more laser diodes. For multi-
wavelength spectrometers a wavelength selection device is used or a series of
optical filters are used for wavelength selection. Wavelength selection devices
include one or more gratings, prisms, and wavelength selective filters. Variation of
the source such as varying which LED or diode is firing is also used for wavelength
selection. Detectors are in the form of one or more single element detectors or one
or more arrays or bundles of detectors. Detector types include InGaAs, PbS, PbSe,
Si, MCT, or the like. Detector arrays include InGaAs, PbS, PbSe, Si, MCT, or the
like. Light collection optics such as fiber optics, lenses, and mirrors are commonly
used in various configurations as an output element within a spectrometer to direct
light from the source to the detector by way of a sample.
Calibration
Glucose concentration analyzers require calibration. This is true for all types of
glucose concentration analyzers such as traditional invasive, alternative invasive,
noninvasive, and implantable analyzers. One fact associated with noninvasive
glucose concentration analyzers is that they are secondary in nature, that is, they do
not measure blood glucose concentrations directly. This means that a primary
method is required to calibrate these devices in order to measure blood glucose
concentrations properly. Many methods of calibration exist.
One noninvasive technology, near-infrared spectroscopy, requires that a
mathematical relationship between an in-vivo near-infrared measurement and the
actual blood glucose concentration is developed. This is achieved through the
collection of in-vivo NIR measurements with corresponding blood glucose
concentrations that have been obtained directly through the use of measurement
tools such as a YSI (YSI INCORPORATED, Yellow Springs, OH) blood glucose
concentration analyzer or any appropriate and accurate traditional invasive reference
device such as the THERASENSE FREESTYLE (THERASENSE, INC., Alameda
CA) glucose concentration analyzer.
For spectrophotometric based analyzers, there are several univariate and
multivariate methods that are used to develop the mathematical relationship between
the measured signal and the actual blood glucose concentration. However, the
basic equation being solved is known as the Beer-Lambert Law. This law states that
the strength of an absorbance/reflectance measurement is proportional to the
concentration of the anaiyte which is being measured, as in Equation 1,
A = εbC (1)
where A is the absorbance/reflectance measurement at a given wavelength of light, ε
is the molar absorptivity associated with the molecule of interest at the same given
wavelength, b is the distance that the light travels in the sample, and C is the
concentration of the molecule of interest (glucose).
Chemometric calibration techniques extract glucose related signal from the
measured spectrum through various methods of signal processing and calibration
including one or more mathematical models. The models are developed through the
process of calibration on the basis of an exemplary set of spectral measurements
known as the calibration set and associated set of reference blood glucose
concentrations based upon an analysis of capillary blood, venous blood, or interstitial
fluid. Common multivariate approaches requiring an exemplary reference glucose
concentration vector for each sample spectrum in a calibration include partial least
squares (PLS) and principal component regression (PCR). Many additional forms of
calibration or optimization are known to those skilled in the art.
An apparatus for measuring infrared throughput typically includes an energy source
emitting infrared energy at multiple wavelengths, an input element, an output
element, and a spectrum analyzer. Tissue is irradiated with multiple wavelengths
from the input element where at least some of the photons are scattered and
absorbed by the tissue. A portion of the photons exit the tissue sample, are
collected by the output element, are directed toward a detector, and are detected.
The resulting signal is utilized in a model for determining the anaiyte concentration.
Calibration Maintenance
Multivariate models reduce a complex measurement in a space modeled with a
reduced number of factors. Data collected for the creation of the original model is
done under a set of conditions. Often this set of conditions changes to the extent
that the original model no longer functions adequately. For example, the
environmental temperature effects the light collection performance of a
spectrometer. In addition to instrumentation and environmental impacts, changes in
the sample affect the model. For example, commonly interference concentrations
vary outside of those tested or new interferences are introduced. In noninvasive
determination of glucose concentration in the body, another key issue with sampling
is that the sample is alive and dynamic in nature. This results in updates to the
calibration being required. >
Prediction
A calibration is used in combination with noninvasive spectra of a subject to
determine the anaiyte concentration of that subject.
Dynamic Properties of Skin
The dynamic properties of skin tissue is an important and largely ignored aspect of
noninvasive glucose determinations. At a given measurement site, skin tissue is
often assumed to remain static, except for changes in the target anaiyte and other
interfering species. However, variations in the physiological state and fluid
distribution of tissue profoundly affect the optical properties of tissue layers and
compartments over a relatively short period of time.
Many factors impact the physical and chemical state of skin. These include
environmental and physiological factors. A list of such factors includes at least body
temperature, environmental temperature, food intake, drug or medicine intake, and
applied pressure to a sampling site. An impact on one part of the body will affect
many other locations in the body. For example, food intake into the digestive track
causes movement of water between internal body compartments. Another example
is caffeine or stimulant intake that changes blood pressure or results in dilation of
capillaries.
NONINVASIVE GLUCOSE DETERMINATION
There exist a number of reports on noninvasive glucose technologies. Some of
these relate to general instrumentation configurations required for noninvasive
glucose determination. Others refer to sampling technologies. Those most related
to the present invention are briefly reviewed here:
General Instrumentation
P. Rolfe, Investigating substances in a patient's bloodstream, UK Patent Application
No. 2,033,575 (August 24, 1979) describe an apparatus for directing light into the
body, detecting attenuated backscattered light, and utilizing the collected signal to
determine glucose concentrations in or near the bloodstream.
C. Dahne, D. Gross, Spectrophotometric method and apparatus for the non-invasive,
U.S. Patent No. 4,655,225 (April 7, 1987) describe a method and apparatus for
directing light into a patient's body, collecting transmitted or backscattered light, and
determining glucose from selected near-IR wavelength bands. Wavelengths include
1560 to 1590, 1750 to 1780, 2085 to 2115, and 2255 to 2285 nm with at least one
additional reference signal from 1000 to 2700 nm.
M. Robinson, K. Ward, R. Eaton, D. Haaland, Method and apparatus for determining
the similarity of a biological anaiyte from a model constructed from known biological
fluids, U.S. Patent No. 4,975,581 (December 4, 1990) describe a method and
apparatus for measuring a concentration of a biological anaiyte such as glucose
using infrared spectroscopy in conjunction with a multivariate model. The
multivariate model is constructed form plural known biological fluid samples.
J. Hall, T. Cadell, Method and device for measuring concentration levels of blood
constituents non-invasively, U.S. Patent No. 5,361,758 (November 8, 1994) describe
a noninvasive device and method for determining anaiyte concentrations within a
living subject utilizing polychromatic light, a wavelength separation device, and an
array detector. The apparatus uses a receptor shaped to accept a fingertip with
means for blocking extraneous light.
R. Barnes, J. Brasch, D. Purdy, W. Lougheed, Non-invasive determination of anaiyte
concentration in body of mammals, U.S. Patent No. 5,379,764 (January 10, 1995)
describe a noninvasive glucose analyzer that uses data pretreatment in conjunction
with a multivariate analysis to determine blood glucose concentrations.
S. Malin, G Khalil, Method and apparatus for multi-spectral analysis of organic blood
analytes in noninvasive infrared spectroscopy, U.S. Patent No. 6,040,578 March 21 ,
2000) describe a method and apparatus for determination of an organic blood
anaiyte using multi-spectral analysis in the near-IR. A plurality of distinct
nonoverlapping regions of wavelengths are incident upon a sample surface, diffusely
reflected radiation is collected, and the anaiyte concentration is determined via
chemometric techniques.
Temperature
It is a well-known that many physiological constituents have near-IR absorbance
spectra that are sensitive in terms of magnitude and location to localized
temperature. This has been reported as impacting noninvasive glucose
determinations, [see K. Hazen, Glucose determination in biological matrices using
near-infrared spectroscopy, Doctoral Dissertation, University of Iowa (August, 1995)].
Coupling Fluid
T. Blank, G. Acosta, M. Mattu, S. Monfre, Fiber optic probe guide placement guide,
U.S. Patent No. 6,415,167 (July 2, 2002) describe a coupling fluid of one or more
perfluoro compounds where a quantity of the coupling fluid is placed at an interface
of the optical probe and measurement site. Perfluoro compounds do not have the
toxicity associated with chlorofluorocarbons.
CALIBRATION ADJUSTMENT
Several methods have been reported to compensate in some part for the dynamic
variation of tissue samples.
One reported method of calibration model generation for noninvasive glucose
concentration determination is to model an individual over a short period of time [see
K. Hazen, Glucose determination in biological matrices using near-infrared
spectroscopy, Doctoral Dissertation, University of Iowa (August, 1995); and J.
Burmeister, In-vitro model for human noninvasive blood glucose measurements,
Doctoral Dissertation, University of Iowa (December 1997)]. This approach avoids
modeling the differences between patients and therefore cannot be generalized to
more individuals. This approach also fails to address the prevalent short-term
problem related to physiologically induced variation and no means of compensating
for variation related to the dynamic water shifts of fluid compartments is reported.
Another approach to overcome the effect of tissue variation on a model is to use
cross-validation. In one study, meal tolerance tests were used to perturb the glucose
concentrations of three subjects and calibration models were constructed specific to
each subject on single days and tested through cross-validation [see Robinson M.R.;
Eaton R.P.; Haaland D.M.; Keep G.W.; Thomas E.V.; Stalled B.R.; and Robinson
P.L. Non-invasive glucose monitoring in diabetic patients: A preliminary evaluation,
Clin Chem 1992;38:1618-22]. This approach models the differences between some
patients presumably with the intent of modeling variations so that future subjects are
predicted by the original model. This approach also fails to address the prevalent
short-term problem related to physiologically induced variation and no means of
compensating for variation related to the dynamic water shifts of fluid compartments
is reported.
Still another approach to overcome the effect of tissue variation on a model is to use
extensive calibration of each subject through a series of glucose perturbations often
over an extended period of time such as 2 to 12 weeks. Often these calibration
periods are followed by an evaluation period during which a subject goes through
one or more additional glucose excursions. The intent is to incorporate into the
model an extensive set of conditions covering future conditions when predictions are
made. When many excursions are used, this incorporation often occurs over a
period of weeks. To date, this extensive calibration and testing protocol has met with
limited success.
Yet another method to overcome the effect of tissue variation on a model is to
compensate for variation related to the structure and state of the tissue through an
intelligent pattern recognition system capable of determining calibration models that
are most appropriate for the patient at the time of measurement [see Malin, S. F.; et.
al. An Intelligent System for Noninvasive Blood Anaiyte Prediction, U.S. Patent
Number 6,280,381]. The calibration models are developed from the spectral
absorbance of a representative population of patients that have been segregated
into groups. The groups or classes are defined on the basis of structural and state
similarity such that the variation within a class is small compared to the variation
between classes. Classification occurs through extracted features of the tissue
absorbance spectrum related to the current patient state and structure.
Still an additional group of approaches to overcome the effect of tissue variation on a
model is calibration transfer. A number of pretreatment of spectral data techniques
have been employed. A general but incomplete list of these pretreatment steps
include trimming, wavelength selection, centering, scaling, normalization, taking an
nth derivative (n>1), smoothing, Fourier transforming, principle component selection,
finite impulse response filtering, linearization, and transformation. This general class
of techniques is found to be limiting in terms of noninvasive glucose concentration
analyzer requirements.
Still an additional approach to overcome the effect of tissue variation on a model is a
group of techniques based upon local centering using a single spectrum [see Lorber
et. al., Local Centering in Multivariate Calibration, Journal of Chemometrics, 1996,
10, 215-220]. In this method, a spectrum is selected for mean centering the
calibration data set that is the closest match (with respect to Mahalanobis distance)
to that of the unknown sample spectrum. A separate partial least squares model is
then constructed for each unknown sample. This technique does not reduce the
spectroscopic variation of the calibration set.
Another approach to overcome the effect of tissue variation on a model is related to
the technique of mean centering [see E. Thomas, R. Rowe, Methods and apparatus
for tailoring spectroscopic calibration models, U.S. Patent No. 6,528,809 (March 4,
2003) and E. Thomas, R. Rowe, Methods and apparatus for tailoring Spectroscopic
Calibration Models, U.S. Patent Number 6,157,041 (December 5, 2000)]. This
method uses spectrographic techniques in conjunction with an improved subject-
tailored calibration model. In calibration data, the model data is modified to reduce
or eliminate subject-specific attributes, resulting in a calibration data set modeling
within-subject physiological variation and instrument variation. In the prediction
phase, the prediction process is modified for each target subject utilizing a minimal
number of spectral measurements for each subject. However, this method does not
address the key problem of short term physiological and chemical changes related to
the dynamic nature of the tissue nor the intra-patient variation related to the
heterogeneity of the tissue sample.
E. Thomas, U.S. 6,528,809, supra and E. Thomas, U.S. 6,157,041 , supra use an
infrared based noninvasive glucose concentration analyzer to obtain absorbance
spectra of human tissue in combination with a model that is periodically corrected
with the use of both a spectrum collected from the tested subject and an invasive
reference glucose concentration determination collected from the tested subject.
The technique employed is loosely referred to as mean centering, though the
subtracted spectrum is not the mean spectrum and the glucose value is used to
correct an offset in the predicted value. Collection of the reference spectrum is time
consuming, requires some expertise on the part of the user, requires data collection
software, and requires a microprocessor or other computing means to implement
into the calibration. In addition, the acquired sample spectrum to be used as a
reference spectrum contains data collection errors and has spectroscopic attributes
not accounted for in the original model. Incorporation of the reference spectrum of
the individual into the model thereby results in a significant potential source of error
in the resulting calibration that directly translates into errors in subsequent glucose
concentration predictions. In a noninvasive glucose determination, this results in an
erroneous glucose concentration being reported that is used as an adjunct method
for directing insulin therapy. For all of these reasons, elimination of the step of
collecting and utilizing a reference sample spectrum is beneficial.
Guide
T. Blank, G. Acosta, M. Mattu, S. Monfre, Fiber optic probe guide placement guide,
U.S. Patent No. 6,415,167 (July 2, 2002) and T. Blank, G. Acosta, M. Mattu, M.
Makarewicz, S. Monfre, A. Lorenz, T. Ruchti, Optical Sampling Interface System For
In Vivo Measurement of Tissue, U.S. patent application number 10/170,921, (filed
June 12, 2002), which are both herein incorporated in their entirety by this reference
thereto, describe use of a guide in conjunction with a noninvasive glucose analyzer
to increase precision of the location of the sampled site resulting in increased
accuracy and precision in a noninvasive glucose concentration determination. The
guide is used for a period of time to increase precision in sampling throughout a
period of sampling, such as a fraction of a day, one day, or a period of multiple days.
Equilibration
A number of reports exist describing the difference (or lack of difference) between
traditional glucose determinations and alternative site glucose determinations. Some
have recognized the potential difference as having impacts upon noninvasive
glucose calibration and maintenance, see U.S. patent application number
10/377,916. The use of heat, rubrifractants, or the application of topical
pharmacologic or vasodilating agents such as nicotinic acid, methyl nicotinamide,
minoxidil, nitroglycerin, histamine, menthol, capsaicin, and mixtures thereof to hasten
the equilibration of the glucose concentration in the blood vessels with that of the
interstitial fluid has been reported, [see Rohrscheib, Mark; Gardner, Craig;
Robinson, Mark R. Method and Apparatus for Non-invasive blood anaiyte
measurement with Fluid Compartment Equilibration, U.S. Patent number 6,240,306,
May 29, 2001 and Robinson, Mark Ries; Messerschmidt, Robert G. Method for Non-
Invasive Blood Anaiyte Measurement with Improved Optical Interface, U.S. patent
number 6,152,876, Nov. 28, 2000].
Release of nitric oxide via photo stimulation is described for use in combination with
noninvasive glucose determinations as a method of equilibrating glucose
concentrations in poorly perfused regions of the body with glucose concentrations of
more well perfused regions of the body. [T. Blank, S. Monfre, M. Makarewicz, M.
Mattu, K. Hazen, and R. Henderson, Photostimulation method and apparatus in
combination with glucose determination, filed May 6, 2004 (attorney docket number
SENS0034).
In all of the related technology of this section, no suggestion of the use of mean
centering utilizing only a reference glucose value in conjunction with a guide that
eliminated the need for a spectral reference is made. Further, no suggestion is
made for easing the use of a bioanalyzer such as a near-IR based noninvasive
glucose analyzer through the use of a guide to reduce the need for mean centering
related techniques based upon spectral references. Further, no minimization of
reference glucose concentration differences has been suggested with the use of
photo stimulation. Finally, to date no FDA device has been approved for the
utilization by an individual or a medical professional for noninvasive glucose
concentration determination.
THE PROBLEM
Physiological parameters that change the state of skin include: tissue hydration, skin
temperature, volume fraction of blood in tissue, skin thickness, magnitude of
absorbance features related to fat, hematocrit concentration, and surface
reflectance. Many of these parameters change over a period of one or more days or
over a much longer period of time such as weeks.
Changes in the state of skin alter a number of properties such as: water
concentration, the concentration of other analytes such as protein, fat, keratinocytes
and glucose, the scattering of skin, the absorbance of skin, the refractive indices of
various layers of skin, the thickness of tissue layers, the emitted radiation from the
body, the mechanical properties of tissue, magnitude of absorbance features related
to water, magnitude of absorbance features related to protein, and the size and
distribution of scattering centers.
Noninvasive spectra, such as a near-IR based diffuse reflectance spectrum, are
representative of skin tissue properties. Since a large number of state changes each
effect a large number of skin tissue properties, variations through time of noninvasive
spectra of a given skin sampling site often vary in a highly nonlinear and profound
manner. Further, factor analysis based multivariate models result in abstract
features. Therefore, change in state often profoundly effects multivariate model
predictions.
Because near-IR based noninvasive glucose analyzers typically use multivariate
analysis that is susceptible to sample state changes, the model must be robust or
optionally updated. The invention herein focuses on maintenance of one or more
calibrations. Calibration maintenance is a costly and time-consuming process that
requires some technical skill. Elimination or automation of steps required for
maintenance of a noninvasive glucose analyzer is beneficial for at least one of
increasing marketability of the analyzer, increasing the number of people who may
use the analyzer, reduction in time requirements associated with a glucose
concentration determination, and increased precision and/or accuracy of a glucose
concentration determination. Specifically, elimination of any data gathering step,
such as collection of spectra, is beneficial for the above reasons. This invention
provides a simple calibration maintenance method for use with a noninvasive
glucose concentration analyzer.
SUMMARY OF THE INVENTION
A method and apparatus for easing the use of an optically based noninvasive
analyzer is presented. More particularly, a simplified algorithm is used that removes
the daily requirement of collecting and using a noninvasive spectrum to update a
calibration model. In another embodiment, a guide is used to substantially reduce
variation in sample probe placement in relation to a skin tissue sampling site,
resulting in the ability to maintain calibration performance with the use of a reference
anaiyte concentration, with or without the use of a reference spectrum collected
nearby in time.
BRIEF DESCRIPTION OF THE DRAWINGS
Figure 1 shows a flat guide coupled to a plug, according to the invention;
Figure 2 presents an LED attachment coupled to a plug with a 4.5 inch radius of
curvature guide, according to the invention;
Figure 3 presents a miniaturized source attachment coupled to a 6.0 inch radius of
curvature guide, according to the invention;
Figure 4 presents a miniaturized source attachment coupled to a flat guide,
according to the invention;
Figure 5 presents spectral variance with and without a guide, according to the
invention; and
Figure 6 presents prediction results overlaid onto a Clarke error grid, according to
the invention.
DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT
The preferred embodiment of the invention uses a near-IR based glucose
concentration analyzer to obtain a spectrum of a sample site of an individual in
conjunction with a model and a reference anaiyte concentration correction to predict
/ measure a glucose concentration of the subject.
As described in E. Thomas, U.S. patent number 6,528,809, supra, and E. Thomas,
U.S. patent number 6,157,041 , supra a traditional offset correction uses both a
reference spectrum from the individual being tested and a direct glucose reading
from the individual. The term reference spectrum herein refers to a tissue spectrum,
such as an initial spectrum of a day, a spectrum from a library, or a matched
spectrum, as opposed to a spectrum of a reference standard, such as an intensity
standard. As discussed above, the collection of the reference spectrum has costs in
terms of instrumentation requirements, software requirements, time, user proficiency,
and is a potential source of error that directly manifests itself in resulting glucose
predictions that is used to direct insulin therapy. For all of these reasons, elimination
of the step of collecting and using a reference sample spectrum is beneficial.
The preferred embodiment provides for a method and apparatus that eliminate the
step of collecting a noninvasive reference spectrum for a periodic update of the
calibration model. That is, neither a reference spectrum of the individual user, a
reference spectrum from a spectral library, nor a matched spectrum to the individual
is used to update a model on a daily, weekly, or monthly basis. Elimination of the
necessity of the reference spectrum is made possible for at least one of the
following reasons:
First, the collection of a reference spectrum does not benefit the net predicted
glucose concentration. Generally, the application of a reference spectrum provides
an offset that is a constant. In addition, the direct reference glucose determination
provides an offset that is a constant. As both offsets are a constant, a single
correction using the direct glucose concentration determination accounts for the
overall error that incorporates both constants. This is detailed below.
In particular, in Equation 1 , XmeaSured is a vector of a prediction spectrum, Wτ is the
vector of coefficients associated with the regression model, and yna, is the resulting
anaiyte (glucose) concentration prediction prior to any correction. Notably, ynat has
an error associated with it. The prediction yhat is thought of as an uncorrected
estimation of the glucose concentration.
hat = (Xmeasured)W (1 )
The estimated ynat glucose concentration prediction has error for a number of
reasons including instrumentation, environmental, and sampling impacts on the
measured spectrum. Of particular importance are tissue volume changes that lead
to a bias through changes in parameters such as the optical pathlength, absorption
coefficient, and scattering coefficient.
In a first method of correcting for the error in the predicted glucose concentration,
yhat that was reported in E. Thomas, U.S. patent number 6,528,809, supra and E.
Thomas U.S. patent number 6,157,041 , supra, a noninvasive reference spectrum,
Xref, is used to localize the measured sample spectrum. The use of the reference
spectrum, Xref, results in an offset to the predicted glucose concentration, ynat error
equal to ybias1, Equation 2. The new partially corrected prediction of the actual
glucose concentration is referred to herein as y'hat Often Xref is a mean spectrum, but
as discussed below there are a number of other sources for the Xref spectrum.
Y'hat = Yhat + Ybiasl = (^measured ~ ^ref) W (2)
Essentially the term XrefWT results in an offset to yhat equal to the constant ybias1. It is
possible that this correction is sufficient to the point the yhat + ybias1 is equal to the
actual glucose concentration within acceptable performance specifications. For
example, the analyzer requirements may have loose accuracy requirements such as
determining if the anaiyte concentration is high or low. This allows correction of the
error to acceptable limits with only the use of a reference spectrum. However,
typically the resulting reference adjusted spectrum after being applied to the model
results in a relative anaiyte (glucose) concentration prediction that still requires an
offset correction. This offset correction is performed using a direct reference glucose
concentration. Typically, a direct reference glucose concentration is a traditional
fingerstick glucose determination, but additional options are discussed below. The
direct glucose concentration reading or reference glucose determination is here
referred to as ybias2. Combining the direct glucose concentration determination step
with the bias correction step of Equation 2 results in a measured anaiyte (glucose)
concentration, ymeas, as in Equation 3.
Ymeas = Yhat + Ybias1 + Ybias2 = (Xmeasured ~ Xref) W + ybjas2 (3)
As indicated above, this first method requires collection and use of both a
noninvasive reference spectrum and a reference glucose concentration.
Going back to Equation 1 , it is again noted that the measured spectrum applied to
the calibration model results in a glucose prediction, yhat, that has an error associated
with it. In a second method, this error is corrected by using only the direct glucose
reference determination without the prior step of subtracting out a mean spectrum.
In this case, the direct reference glucose determination here referred to as yΘrr
provides an estimated error of the predicted glucose concentration yhaf, Equation 4
used to estimate a glucose concentration.
Ymeas = Yhat + Yβrr = XmeasuredWT + yerr (4)
Notably, the estimated error, yΘrr, using this second method is seen by comparison of
Equations 3 and 4 to represent both of the errors from the more complicated 2-step
method above that resulted in the ybias1 and ybias2 terms.
An alternative embodiment of the invention is now presented that uses one or more
stored reference spectra. In the methods above, it is common to refer to the
reference spectrum as a mean spectrum because a common technique is to
subtract out the spectrum, and this is often done with a mean spectrum. However,
the reference spectrum is one of a number of spectra, including the first spectrum of
a time period where the time period includes a day, week, month, or even a
spectrum collected for an individual upon delivery of the instrument. Typically, the
correction is performed once a day, waking day, or major fraction of a day. Instead
of using the first spectrum of a time period, the average of the first n spectra of a
time period is used as may a spectrum that is a linear combination of any number of
time adjacent or time separated spectra of a time period. Optionally, the reference
spectra originates from a data base such as a spectral library and is chosen based
upon spectral features or based upon a calculation such as a Mahalanobis distance
calculation. In an additional embodiment, a basis set gold standard spectrum, such
as a spectrum of water, is used as the reference spectrum. The advantage of a fixed
gold standard spectrum is that it is collected in a controlled environment and is
digitally stored in the analyzer for use in subsequent model updates.
Still an additional embodiment of the invention uses a near-IR based glucose
concentration analyzer to obtain a spectrum of a sample site of an individual in
conjunction with a guide, a model, and a reference anaiyte concentration to predict /
measure a glucose concentration of the individual.
Spectrometer light throughput is effected by a number of factors, including
environmental states and the state of a sample. In addition, changes in
environmental states and changes in the state of the body result in dynamic changes
to a living tissue sample site. Examples of change that alter the tissue state include
temperature changes and/or distributions, localized pressure changes or
distributions, water movement, glucose movement, and changes in body
composition such as protein, fat, and hematocrit. These changes affect the chemical
and physical attributes of the body. Physical changes include localized and/or
regional changes in the refractive index, absorption coefficient, anisotropy, and
scattering coefficient that affect probing photon penetration, radial transport, and
optical pathlength. Combined, these state changes result in a high-degree of
nonlinearity being observed in noninvasive spectra. Many of these changes have a
large dependency on the positioning of an optical sampling probe, that includes
excitation and/or collection elements of a spectrometer, relative to the sample
volume.
The physiology of skin is not homogeneous. The chemical and physical make-up of
skin depends upon the position of the skin. Many differences in skin structure with
position exist, including thickness of skin layers and the skin physical constituents.
Noninvasive spectra of the body are dependent upon the sample site. For example,
the spectra of a forearm skin tissue sample is different depending upon where on the
forearm the skin is sampled. For example, a skin sample varies as a function of
position in terms of constituents such as water, fat, and protein and in properties
such as the scattering coefficient.
Guide
In its broadest sense, a guide is an element that limits positioning of a sample probe
relative to a sample site. The guide couples an input and output element to a
targeted tissue volume, thereby controlling and reducing spectral variation, and
results in decreased error in a determined anaiyte concentration. A guide is
illustrated as one-half of a lock and key mechanism where the guide lock limits the
positioning of the sample probe (key) relative to the sample site. A guide aids in the
reproducibility of locating sampling and collection elements, such as a sample probe
and fiber optics of a noninvasive spectrometer based analyzer, in relation to a
sampling site and is beneficial to a noninvasive anaiyte concentration determination
due to the associated reduction of nonlinear variability. A guide has been taught in
U.S. Patent No. 6,415,167 (July 2, 2002), and U.S. Patent Application No.
10/170,921 (filed June 12, 2002), which are both herein incorporated in their entirety
by this reference thereto.
A number of guide and attachment apparatus are described herein. Preferably, the
attachments have the same interface so that a single guide element is used with
each attachment. Similarly, it is preferable to have each of the pieces of apparatus
that attach to an attachment have the same interface. For example, the guide and a
reference sample preferably have the same interface so that they both couple to a
sample probe. A common interface allows any of the guides or a reference to
interface with any of the attachments, such as the plug, photonic stimulator, sample
module, or miniaturized source.
Lock (Guide)
A sample site varies between individuals in terms of circumference or radius of
curvature. For body parts, such as an abdominal region or a finger, the radius of
curvature may range from flat to 0.375 inch, respectively. Even within a body part,
the radius of curvature varies between individuals. For example, some individuals
have small diameter arms while others have larger diameter arms. Matching the
shape of a guide to the structure of the sample site results in increased precision of
subsequent optical sampling.
Examples of guides are those that have flat sample interfaces and those with a 6.0
inch, 4.5 inch, and 3.0 inch radius of curvature. Figure 1 presents a guide element
with a flat radius of curvature. For the case of an arm sampling site, the skinnier the
arm the smaller the radius of curvature of the optimal guide. The guide presented in
Figure 1 is presented in relation to a plug, which is one of the guide attachments
discussed below. A core feature of the guide element is that is makes up one-half of
a lock and key combination. That is, a surface exists that reproducibly guides the
other half of a lock and key element into a reproducible position. In this case, the
lock element is in the guide, but alternatively it is in the attachment. In this case, the
lock element is a hole in the guide that is roughly rectangular with two opposing
sides each having rounded shapes. The rectangular shape limits rotational
alignment. Preferably, the guide would not have rotational freedom. For instance,
the pictured guide has C2 symmetry allowing it to be rotated by 180 degrees and still
having the same shape. This rotational freedom is obtained by many means, such
as by flattening one of the round ends. In a second case, the lock element has pins
extending from it that interface into associated holes in the key (attachment) element
to position the key relative to the guide reproducibly.
Many lock element shapes are readily used. Examples include virtually any
geometrically shaped hole or any shape (not necessarily a hole) that provides
reproducible positioning while, preferably, preventing freedom of rotation. In the
particular guide elements presented, optional additional holes or divots are pictured.
The function of these is primarily to reduce weight, minimize surface abnormalities,
such as sink marks on the sampling site, and to maintain strength while limiting the
twisting freedom of the guide. An additional optional component pictured on these
guides are magnets. The magnets are used to control contact force and/or to aid in
alignment of the lock and key mechanism. In the guide pictured, optional opposing
pole magnets are also placed into the plug. Of the paired magnets, one half of the
pair is optionally a metallic substance, such as sheet metal or stainless steel, which
reduces cost and/or weight. Many additional mechanical structures and will be
obvious to those skill in the art that allow the lock element to interface with the guide
element
The guide is attached to a sample site with a device, such as a band, strap, hook
and loop technology, or preferentially with a double sided adhesive. Commonly, the
adhesive is firmly placed onto the sample site and then the guide is visually aligned
onto the adhesive. This sequence reduces separation events of the adhesive from
the sample site. Optionally, the adhesive is attached to the guide and the pair is
placed into contact with the sample site as a unit. This eases alignment of the guide
to the adhesive. Optionally, the adhesive comes to the user already attached to the
guide element. The guide and adhesive are semi-permanently and removeably
attached to the sampling site. The guide is typically left in place for the remainder of
a sampling period such as one waking day or the length of a data collection period,
such as 2, 4, or 8 hours.
An optional intermediate layer or guide extension is used between the guide and the
double sided adhesive that attaches to the sampling site. Essentially, this is a semi-
flexible material such as acetate. The material allows some flexibility to allow the
sample site skin to stretch. This reduces sampling transients resulting from
movement of the subject. Conversely, in subjects with poor turgor, the skin flexes
too much and a more rigid insert, such as a plastic film is used.
The guide is preferentially formed out of a thermoplastic, such as a polycarbonate or
a polyurethane. However, many materials will be obvious to those skilled in the art.
Because the guide is in contact with the sampling site (sometimes with an
intermediate adhesive), the thermal properties of the guide become important.
Typically, the guide is non-thermally conductive to reduce sampling site temperature
gradients. However, in some cases a thermally conductive guide is preferential,
such as when heat flow to or from the sample site is desired. The guide material is
preferentially biocompatible.
Preferentially, the guide is optically coupled to the sampling site through the use of
an index of refraction matching medium such as a fluoropolymer, a fluorocompound,
Fluorinert, FC-40, FC-70, or equivalent.
Key (Attachment)
The other half of the guide lock and key mechanism is herein referred to as an
attachment to the guide element. Examples of attachments to the guide include a
plug, a photonic stimulator, a miniaturized source, and the tip of a sample module of
a spectrometer. An example of each of these attachments is provided below.
A plug attachment is presented in Figure 1 coupled to a guide. The plug functions to
accomplish at least one of hydration of the sampling site by occlusion, protection of
the sampling site from physical perturbation, protection of the sampling site from
contamination, alignment of the guide, and allowing an aesthetic appearance, such
as a watch, ring, or graphical symbol. The pictured plug has two protruding
elements with a cross support designed for ease of gripping. Optionally, the plug is
made to look or function like a wristwatch that may or may not have a band that
looks like a watchband with it. In an alternative embodiment, the plug is miniaturized
to resemble a decorative object, such as a ring.
The pictured plug in Figure 1 has an optional central tunnel. This tunnel is used in
the initial placement of the guide. In this method, a double sided adhesive strip is
attached to the sampling site. The adhesive strip has an opening in it that is slightly
larger than the optical probing element. After an adhesive is placed onto the arm,
the guide is attached to the plug and slid down a guiding rod to the adhesive so that
the optical path is centered in the cutout on the adhesive. Essentially, the rod
through the plug and guide is used as a sighting mechanism.
A photonic stimulator attachment is presented in Figure 2 coupled to a guide with a
4.5 inch radius of curvature sample interface. Photo-stimulation at or near at least
one sample site is used to enhance perfusion of the sample site leading to reduced
errors associated with sampling. Increased perfusion of the sample site leads to
increased volume percentages of the target anaiyte and/or allows the blood or tissue
constituent concentrations to more accurately and/or precisely track corresponding
sample constituents in more well perfused body compartments or sites, such as
arteries, veins, or fingertips. In one embodiment, analysis of the photo-stimulated
site is used in conjunction with glucose analyzers to determine the glucose anaiyte
concentration with greater ease, accuracy, or precision and allows determination of
the anaiyte concentration of another non-sampled body part or compartment. This
technology is described in a U.S. patent application (attorney docket number
SENS0034) and is herein incorporated in its entirety by this reference thereto.
A sample module is presented in Figure 3 coupled to a guide with a 6.0 inch radius
of curvature sample interface. Sample modules are described in U.S. Patent
application No. 10/472,856 filed September 18, 2003 (attorney docket number
SENS0011). A sample module is preferably part of an analyzer, the analyzer
additionally comprising a base module and a communication bundle. The sample
module is attached continuously or semi-continuously to a human subject and
collects spectral measurements that are used to determine a biological parameter in
the sampled tissue. The preferred target anaiyte is glucose. The preferred analyzer
is a near-IR based glucose analyzer for determining the glucose concentration in the
body.
A miniaturized source attachment is presented in Figure 4 in combination with a
reference guide element. A reference element is used to provide a signal
representative of the current state of the analyzer. Preferably, the reference element
is contained in an element that has an interface that is functionally identical to the
guide interface so that the reference couples to a sample module or a miniaturized
source in the same manner as the guide.
Guide Data
Standard deviations of a series of near-IR noninvasive spectra of an arm of a single
individual are presented in Figure 5 for spectra collected with a guide 501 and
without a guide 502. The standard deviation in absorbance units are observed to be
much smaller with the use of a guide. Those skilled in the art will recognize the
deviations as dominated by variations due to pathlength and the probed water
concentration with smaller variations being related to protein, fat, and temperature.
Hence, the guide increases the precision of locating the optical probe in relation to
the sample site that in turn, due to reduction in sample variation with position, result
in a smaller variation in the spectral readings.
Example
An example data set of noninvasive glucose concentration prediction is presented in
Figure 6. These representative data are processed using the preferred model
correction technique described above that uses only a direct glucose reference
concentration and does not use a noninvasive (or related) reference spectrum. The
instrumentation used to collect the calibration data includes three identically
configured near-IR based analyzers configured with a tungsten halogen source,
excitation guiding optics, a bandpass filter, collection guiding optics, a slit, a grating,
an array detector, and associated electronics. A guide is placed onto each subject at
the beginning of a given day for all measurements taken in the calibration and
subsequent prediction measurements. All subjects underwent one or more glucose
profiles over a period of at least one day. The calibration is built with a total of eight
subjects, using the calibration instruments to collect 241 samples over a period of
two months. A model is built with the data utilizing spectra from 1200 to 1800 nm
preprocessed with a 27-point Savitsky-Golay first derivative. Outlier detection
routines include modules for detecting and eliminating samples with poor surface
contact between the optics and the sampled site, samples with large tissue
transients, and samples with undue tissue distortion. A multivariate partial least
squares model is used. Prediction spectra are collected with a similar glucose
analyzer with differences only in the mechanism that holds the excitation and
collection optics. The prediction spectra were collected over a five week period
initiated seven months after the end of collection of the calibration data. The
prediction data set included spectra collected from twenty-one test subjects on three
instruments resulting in 368 prediction spectra. The model generated with the
calibration data set was employed without modification in a blind fashion. An offset
correction was used that employed a glucose reference concentration collected with
the first sample of a given day. No reference spectrum from any of the prediction
spectra were used in the prediction, nor was any noninvasive spectrum or
synthesized noninvasive spectrum used in any phase of this prediction data set.
Again, a reference glucose concentration is used to update the model for
subsequent prediction of a glucose concentration with a noninvasive analyzer
without the use of a corresponding noninvasive reference spectrum. The resulting
predictions presented in Figure 6 are overlaid onto a Clarke error grid. A total of
98.7% of the resulting glucose concentration predictions fell within the clinically
acceptable 'A' or 'B' region of the Clarke error grid, and the resulting standard error
of prediction was 26.3 mg/dL. This clearly demonstrates efficacy of the analyzer.
Notably, the prediction performance is virtually identical to the calibration
performance that resulted in 100% of the glucose concentration determinations
falling within the 'A' or 'B' region of the Clarke error grid and had a standard error of
calibration of 27.0 mg/dL.
In the methods described herein, it is common to refer to a reference spectrum as a
mean spectrum because a common technique is to subtract out the spectrum and
this is often done with a mean spectrum. However, an optional reference spectrum
includes one of a number of spectra, including the first spectrum of a time period
where the time period includes a day, week, month, or even a spectrum collected for
an individual upon delivery of the instrument. Typically, the correction is performed
once a day, waking day, or major fraction of a day. Alternatively, the removed
spectrum is the first spectrum of a time period, the average of the first n spectra of a
time period is used, or a spectrum that is a linear combination of any number of time
adjacent or time separated spectra of a time period. In yet another embodiment, the
reference spectrum is from a data base, such as a spectral library, and is chosen
based upon spectral features or based upon a calculation, such as a Mahalanobis
distance calculation. For example, a gold spectrum is used as described above.
In the methods above, the direct reference anaiyte concentration is typically a
capillary based blood glucose determination performed with traditional enzymatic,
electro-enzymatic, or colorimetric means. However, the reference glucose
concentration is alternatively generated by a minimally invasive glucose meter or a
noninvasive meter. Generally, an FDA approved reference method is used.
Although the invention is described herein with reference linear models, it is
recognized that a non-linear model, such as a neural network, is also applicable to
anaiyte concentration estimation.
Although the invention is described herein with reference to the preferred
embodiments, one skilled in the art will readily appreciate that other applications may
be substituted for those set forth herein without departing from the spirit and scope of
the present invention. Accordingly, the invention should only be limited by the
Claims included below.