U.S. patent application number 13/166806 was filed with the patent office on 2011-12-22 for health monitoring system.
Invention is credited to Eyal Dassau, Francis J. Doyle, III, Rebecca A. Harvey, Howard Zisser.
Application Number | 20110313680 13/166806 |
Document ID | / |
Family ID | 45329398 |
Filed Date | 2011-12-22 |
United States Patent
Application |
20110313680 |
Kind Code |
A1 |
Doyle, III; Francis J. ; et
al. |
December 22, 2011 |
Health Monitoring System
Abstract
A machine for processing continuous glucose monitoring data and
issuing an alert if hypoglycemia is imminent has three modules: (a)
a pre-processing module that receives and modulates continuous
glucose monitoring data by reducing noise and adjusting for missed
data points and shifts due to calibration; (b) a core algorithm
module that receives data from the pre-processing module and
calculates a rate of change to make a hypoglycemia prediction, and
determine if hypoglycemia is imminent; and (c) an alarm mode module
that receives data from the core algorithm and issues an audio or
visual alert or warning message or a negative feedback signal to an
insulin delivery device if hypoglycemia is imminent.
Inventors: |
Doyle, III; Francis J.;
(Santa Barbara, CA) ; Dassau; Eyal; (Goleta,
CA) ; Zisser; Howard; (Santa Barbara, CA) ;
Harvey; Rebecca A.; (Goleta, CA) |
Family ID: |
45329398 |
Appl. No.: |
13/166806 |
Filed: |
June 22, 2011 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
61357409 |
Jun 22, 2010 |
|
|
|
Current U.S.
Class: |
702/19 |
Current CPC
Class: |
A61B 5/7275 20130101;
A61B 5/746 20130101; G16H 40/67 20180101; G16H 50/20 20180101; A61B
5/14532 20130101; G16H 40/63 20180101; G16H 20/60 20180101; G16H
15/00 20180101 |
Class at
Publication: |
702/19 |
International
Class: |
G06F 19/00 20110101
G06F019/00 |
Goverment Interests
[0002] This work was supported by grant ROI-DK085628-01 from the
National Institutes of Health; the Government has certain rights in
this invention.
Claims
1. A low glucose prediction signal generator that uses a set of
constraints to predict an imminent occurrence of hypoglycemia, the
generator comprising: (a) a pre-processing module that receives and
modulates continuous glucose monitoring (CGM) data by reducing
noise and adjusting for missed data points and shifts due to
calibration; (b) a core algorithm module that receives data from
the pre-processing module and calculates a rate of change to make a
hypoglycemia prediction, and determine if hypoglycemia is imminent;
and (c) an alarm mode module that receives data from the core
algorithm and if hypoglycemia is imminent, issues an audio or
visual alert or warning message or a negative feedback signal to an
insulin delivery device.
2. The signal generator of claim 1 wherein the preprocessing module
the CGM data are filtered for noise using a noise spike filter to
remove outliers and a low pass filter to damp electrical noise; to
use most current information, recently missed data points are
interpolated using a simple linear interpolation; to prevent
erroneous estimation of the rate of change when the sensor is
calibrated, a calibration detection module is used to detect a
persistent offset in data and shifts the data from before the
calibration; wherein the preprocessing module only operates when
enough data is present to make a prediction and will operate during
periods of sensor outage, up to two readings, by extrapolating
previous estimates.
3. The signal generator of claim 1 wherein the core algorithm
module the rate of change is estimated using the first derivative
of the 3-point Lagrange interpolation polynomial, wherein a series
of logical steps is taken to ensure that the subject is within a
determined proximity of the hypoglycemia threshold, the glucose is
decreasing at a physiologically probable rate, and that the time to
crossing the hypoglycemia threshold is within a preset prediction
horizon, and wherein if these checkpoints are all passed, the alarm
mode module is activated.
4. The signal generator of claim 1 wherein the alarm mode module,
when an imminent hypoglycemic event is predicted in the core
algorithm module, the alarm mode references any previous alarms to
ensure that it has been more than a pre-designated lockout period
to ensure that any action taken during the previous alarm has time
to take effect, wherein if this checkpoint is passed, an audible,
electronic or visible alarm is issued, or a feedback signal is
issued that results in insulin delivery suspension, insulin
delivery attenuation, or consumption of rescue carbohydrates.
5. The signal generator of claim 1 wherein: (a) the preprocessing
module the CGM data are filtered for noise using a noise spike
filter to remove outliers and a low pass filter to damp electrical
noise; to use most current information, recently missed data points
are interpolated using a simple linear interpolation; to prevent
erroneous estimation of the rate of change when the sensor is
calibrated, a calibration detection module is used to detect a
persistent offset in data and shifts the data from before the
calibration; wherein the preprocessing module only operates when
enough data is present to make a prediction and will operate during
periods of sensor outage, up to two readings, by extrapolating
previous estimates; (b) the core algorithm module the rate of
change is estimated using the first derivative of the 3-point
Lagrange interpolation polynomial, wherein a series of logical
steps is taken to ensure that the subject is within a determined
proximity of the hypoglycemia threshold, the glucose is decreasing
at a physiologically probable rate, and that the time to crossing
the hypoglycemia threshold is within a preset prediction horizon,
and wherein if these checkpoints are all passed, the alarm mode
module is activated; and (c) the alarm mode module, when an
imminent hypoglycemic event is predicted in the core algorithm
module, the alarm mode references any previous alarms to ensure
that it has been more than a pre-designated lockout period to
ensure that any action taken during the previous alarm has time to
take effect, wherein if this checkpoint is passed, an audible,
electronic or visible alarm is issued, or a feedback signal is
issued that results in insulin delivery suspension, insulin
delivery attenuation, or consumption of rescue carbohydrates.
6. The signal generator of claim 1 wherein the preprocessing module
implements the steps of FIGS. 1-1 and 1-2.
7. The signal generator of claim 1 wherein the core algorithm
module implements the steps of FIG. 2.
8. The signal generator of claim 1 wherein the alarm mode module
implements the steps of FIG. 3-3.
9. The signal generator of claim 1 wherein the preprocessing module
implements the steps of FIGS. 1-1 and 1-2, the core algorithm
module implements the steps of FIG. 2-1, and the alarm mode module
implements the steps of FIG. 3-3.
10. A machine for processing continuous glucose monitoring (CGM)
data and issuing an alert if hypoglycemia is imminent, the machine
comprising a computer specifically programmed with: (a) a
pre-processing module that receives and modulates continuous
glucose monitoring (CGM) data by reducing noise and adjusting for
missed data points and shifts due to calibration; (b) a core
algorithm module that receives data from the pre-processing module
and calculates a rate of change to make a hypoglycemia prediction,
and determine if hypoglycemia is imminent; and (c) an alarm mode
module that receives data from the core algorithm and issues an
audio or visual alert or warning message or a negative feedback
signal to an insulin delivery device if hypoglycemia is
imminent.
11. The machine of claim 10 operably-linked to the insulin delivery
device.
12. The machine of claim 10, operably-linked to a continuous
glucose monitoring (CGM) device.
13. The machine of claim 10 operably-linked to a integrated
continuous glucose monitoring (CGM) and insulin delivery
device.
14. A method of using a machine of claim 10 for processing
continuous glucose monitoring (CGM) data and issuing an alert if
hypoglycemia is imminent, the method comprising the steps of: (a)
receiving and modulating CGM data in a pre-processing module by
reducing noise and adjusting for missed data points and shifts due
to calibration; (b) receiving data from the pre-processing module
in a core algorithm module that then calculates a rate of change to
make a hypoglycemia prediction, and determine if hypoglycemia is
imminent; and (c) receiving data from the core algorithm in an
alarm mode module that then issues an audio or visual alert or
warning message or a negative feedback signal to an insulin
delivery device if hypoglycemia is imminent.
15. A low glucose predictor (LPG) core algorithm comprising a
numerical logical algorithm that feeds a three-point calculated
rate of change using backward difference approximation and the
current glucose value into logical expressions to detect impending
hypoglycemia, wherein the logical expressions verify that the rate
of change is both negative and within a predetermined acceptable
range as well as that the continuous glucose monitoring (CGM)
glucose values are within predefined boundaries and that a pending
hypoglycemic event is predicted within the threshold time window,
and wherein the numerical logical algorithm provides tuning and
insensitivity to sensor signal dropouts.
Description
CROSS REFERENCE TO RELATED APPLICATION
[0001] This application claims priority under 35 U.S.C. .sctn.119
from Provisional Application Ser. No. 61/357,409, filed Jun. 22,
2010, the disclosure of which is incorporated herein by
reference.
FIELD OF THE INVENTION
[0003] The field of the invention is a continuous glucose
monitoring.
BACKGROUND OF THE INVENTION
[0004] Diabetes is a chronic disease only controlled by constant
vigilance. Chronic elevations, and likely fluctuations, of the
blood glucose may result in long term complications (blindness,
kidney failure, heart disease, and lower extremity amputations).
Perversely, attempts at normalizing glucose concentrations also
increases the risk of serious health issues related to
hypoglycemia. Despite the use of insulin infusion pumps and
programs that promote intensive diabetes management, the average
A1c (an indicator of long-term blood glucose control) reported by
major diabetes treatment centers remains higher than 8%, well above
the recommended goal of 6.5-7%. Many factors contribute to this
failure:
1) the difficulties in correctly estimating the amount of
carbohydrates in a meal, 2) missed meal boluses, and 3) anxiety
about anticipated hypoglycemia, resulting in patients giving
themselves less insulin, especially overnight.
[0005] It has always been difficult to achieve compliance with
complicated medical regimens, such as the administration of insulin
three or more times a day. As long as diabetes treatment demands
constant direct intervention, the vast majority of people with
diabetes will not meet treatment goals. An expanding area of
research addressing diabetes is working on developing automated
closed loop systems that integrates glucose readings and insulin
delivery without the on-going active intervention of the
patient--an "artificial pancreas".
[0006] We have developed an automated closed-loop system that
contains a subcutaneous continuous glucose monitor and a
subcutaneous insulin delivery pump for type 1 diabetes patients.
These two components are connected by a control algorithm using
data from the glucose sensor to determine the appropriate insulin
delivery. We use a health monitoring system (HMS) algorithm that
adds an independent safety layer to the overall system. The HMS
analyzes CGM data and CGM trends in anticipation of impending
hypoglycemia. The HMS issues electronic, visual and/or audio alerts
in response to impending hypoglycemia (e.g. within 15 minutes),
such as on the AP device screen, with a request for the
investigator to intervene and treat the subject, e.g. with 16 g
carbohydrate. A secondary alert may be sent as a text message, such
as to the clinical team, that hypoglycemia is predicted and may
also suggest taking outside action, such as eating carbohydrates,
in order to prevent hypoglycemia.
RELEVANT LITERATURE
[0007] Dassau E., F. Cameron, H. Lee, B. W. Bequette, H. Zisser, L.
Jovanovi{hacek over (c)}, H. P. Chase, D. M. Wilson, B. A.
Buckingham, and F. J. Doyle. Real-Time Hypoglycemia Prediction
Suite Using Continuous Glucose Monitoring: A Safety Net for the
Artificial Pancreas. Diabetes Care, 33(6):1249-1254, 2010. [0008]
Dunn T. C., R. C. Eastman, and J. A. Tamada. Rates of Glucose
Change Measured by Blood Glucose Meter and the GlucoWatch
Biographer During Day, Night, and Around Mealtimes. Diabetes Care,
27(9):2161-2165, 2004. [0009] Seborg D. E., T. F. Edgar, D. A.
Mellichamp, and F. J. Doyle III, Process Dynamics and Control, 3rd
ed., Hoboken, N.J.: John Wiley & Sons, 2011. [0010] Buckingham
B, Cobry E, Clinton P, Gage V, Caswell K, Kunselman E, Cameron F,
Chase H P. Preventing hypoglycemia using predictive alarm
algorithms and insulin pump suspension. Diabetes Technol Ther 2009;
11:93-97
SUMMARY OF THE INVENTION
[0011] The invention provides computer-implemented algorithms,
computers programmed with a subject algorithm, and methods and
machines for processing continuous glucose monitoring (CGM) data
and issuing an alert or negative feedback signal if hypoglycemia is
imminent.
[0012] In one embodiment the invention provides a low glucose
predictor and signal generator that uses a set of constraints to
predict an imminent occurrence of hypoglycemia, comprising: (a) a
pre-processing module that receives and modulates continuous
glucose monitoring (CGM) data by reducing noise and adjusting for
missed data points and shifts due to calibration; (b) a core
algorithm module that receives data from the pre-processing module
and calculates a rate of change to make a hypoglycemia prediction,
and determine if hypoglycemia is imminent; and (c) an alarm mode
module that receives data from the core algorithm and if
hypoglycemia is imminent, issues an audio or visual alert or
warning message or a negative feedback signal to an insulin
delivery device.
[0013] In another embodiment of the invention provides a machine
for processing continuous glucose monitoring (CGM) data and issuing
an alert if hypoglycemia is imminent, the machine comprising a
computer specifically programmed with: (a) a pre-processing module
that receives and modulates continuous glucose monitoring (CGM)
data by reducing noise and adjusting for missed data points and
shifts due to calibration; (b) a core algorithm module that
receives data from the pre-processing module and calculates a rate
of change to make a hypoglycemia prediction, and determine if
hypoglycemia is imminent; and (c) an alarm mode module that
receives data from the core algorithm and issues an audio or visual
alert or warning message or a negative feedback signal to an
insulin delivery device if hypoglycemia is imminent.
[0014] In another embodiment the invention provides a low glucose
predictor (LPG) core algorithm comprising a numerical logical
algorithm that feeds a three-point calculated rate of change using
backward difference approximation and the current glucose value
into logical expressions to detect impending hypoglycemia, wherein
the logical expressions verify that the rate of change is both
negative and within a predetermined acceptable range as well as
that the continuous glucose monitoring (CGM) glucose values are
within predefined boundaries and that a pending hypoglycemic event
is predicted within the threshold time window, and wherein the
numerical logical algorithm provides tuning and insensitivity to
sensor signal dropouts.
[0015] In another embodiment the invention provides a method of
using a subject machine, programmed-computer or algorithm for
processing continuous glucose monitoring (CGM) data and issuing an
alert or signal if hypoglycemia is imminent, the method comprising
the steps of: (a) receiving and modulating CGM data in a
pre-processing module by reducing noise and adjusting for missed
data points and shifts due to calibration; (b) receiving data from
the pre-processing module in a core algorithm module that then
calculates a rate of change to make a hypoglycemia prediction, and
determine if hypoglycemia is imminent; and (c) receiving data from
the core algorithm in an alarm mode module that then issues an
audio or visual alert or warning message or a negative feedback
signal to an insulin delivery device if hypoglycemia is
imminent.
[0016] In particular embodiments of the subject inventions, in the
preprocessing module the CGM data are filtered for noise using a
noise spike filter to remove outliers and a low pass filter to damp
electrical noise. To use the most current information, recently
missed data points are interpolated using a simple linear
interpolation. To prevent erroneous estimation of the rate of
change when the sensor is calibrated, a calibration detection
module is used: this detects a persistent offset in data and shifts
the data from before the calibration accordingly. This module only
operates when enough data is present to make a prediction (number
of points required denoted as PR). If the number of points is less
than PR or there are large gaps in the last PR points this module
will not operate. This will operate during periods of sensor outage
(up to two readings) by extrapolating previous estimates.
[0017] In particular embodiments of the subject inventions, in the
core algorithm module the rate of change is estimated using the
first derivative of the 3-point Lagrange interpolation polynomial.
A series of logical steps is taken to ensure that the subject is
within a reasonable proximity of the hypoglycemia threshold, the
glucose is decreasing at a physiologically probable rate, and that
the time to crossing the hypoglycemia threshold is within a preset
prediction horizon. If these checkpoints are all passed, the alarm
mode module is activated.
[0018] In particular embodiments of the subject inventions, in the
alarm mode module when an imminent hypoglycemic event is predicted,
the alarm mode references any previous alarms to ensure that it has
been more than a pre-designated lockout period. This is to ensure
that any action taken during the previous alarm has time to take
effect. If this checkpoint is passed, an audible, electronic and
visible alarm is issued. Methods of action may be any of the
following: insulin delivery suspension, insulin delivery
attenuation, or consumption of rescue carbohydrates.
[0019] In another particular embodiment the preprocessing module
implements the steps of FIGS. 1-1 and 1-2, the core algorithm
module implements the steps of FIG. 2, and/or the alarm mode module
implements the steps of FIG. 3-3.
[0020] In particular embodiments the subject inventions are
operably-linked to an insulin delivery device and/or to a
continuous glucose monitoring (CGM) device.
[0021] The invention provides all combinations of the recited
particular embodiments as if each combination had been separately
recited.
BRIEF DESCRIPTION OF THE DRAWINGS
[0022] FIG. 1-1. Flow chart of the Pre-Processing module.
[0023] FIG. 1-2. Flow chart of the Missed Point Handling section of
the Pre-Processing module.
[0024] FIG. 2. Flow chart of the Core Algorithm module.
[0025] FIG. 3-1. Screenshot of the impending hypoglycemia pop-up
window.
[0026] FIG. 3-2. Representation of the message when the CGM is
below 70 mg/dL.
[0027] FIG. 3-3. Flow chart of the Alarm Mode module.
[0028] FIG. 4. Control algorithms overview.
[0029] FIG. 5. Text message view to inform of predicted
hypoglycemia
DETAILED DESCRIPTION OF SPECIFIC EMBODIMENTS OF THE INVENTION
[0030] The HMS functions as a process monitoring module that is
executed in real time. This section of the control algorithm serves
as a safety layer to the device. The zone-MPC algorithm controls
the delivery of insulin, while the HMS evaluates the trend of the
glucose in a different way in order to provide an extra layer of
safety to ensure the health of the subject. The HMS will generate
an audible and visual alert to the clinicians and send a text
message to the physician in charge with a profile of the current
trend and prediction for the upcoming 15 minutes. The key module of
the HMS is the low glucose predictor (LGP) that uses a set of
constraints to predict the imminent occurrence of hypoglycemia. The
relevant variable is glucose concentration, G, assumed to be the
CGM measurement. The LGP has three major modules: a pre-processing
module to get the CGM data ready for prediction; a core algorithm
section to calculate the rate of change, make predictions, and
determine if hypoglycemia is imminent; and an alarm mode module to
issue the audible and visual alert and send the warning text
message.
[0031] The HMS can work with a control algorithm or without one,
and any control algorithm can be used to deliver insulin. Insulin
can be also delivered manually by the user. In addition, HMS
parameters can be adjusted. e.g. PH 10-60 min, THactivation 90-150
mg/dL, THhypo 60-80 mg/dL, LT 15-45 min.
1. Pre-Processing
[0032] The pre-processing module is used to get the CGM data ready
for prediction. The CGM often has noisy data, missed data points,
and shifts due to calibration. These issues are all addressed in
the pre-processing module. A flow diagram of the module can be seen
in FIG. 1-1, with terms detailed in Table 1-1.
1.1 Missed Point Handling
[0033] The HMS is called every five minutes regardless of missing
data. To avoid missing a hypoglycemic event when G is low and
falling and data is missing, the HMS will function when up to two
points are missing. The estimation of the rate of change from the
previous point is projected for the missing data (these data are
not saved, only used for current hypoglycemia alarming if
necessary). The HMS then proceeds directly to the Core Algorithm
module using the predicted data as G(j) where j=k for one missed
point and j=k-1, k for two missed points. Here,
G.epsilon..sup.k.times.1. A flow diagram of this branch of the
pre-processing module can be seen in FIG. 1-2.
1.2 Shift Detection
[0034] When the CGM is calibrated, a shift in the CGM data is
introduced. In order to make a more accurate prediction, these
shifts must be detected so that the shift does not produce a
non-physiologic rate of change calculation. A shift in the signal
is detected when the change in the raw signal is large (>4
mg/dL/min, considered to be non-physiologic) and then the next
point continues roughly the same trend as before the shift, but
with an offset (Dunn et al., 2004, supra). When a shift is
detected, the points after the shift can be considered to be more
accurate, and the same offset can be applied to the points before
the shift to reflect the true trend. If a shift is detected, the
previous points are shifted as follows:
Shift detected if .DELTA. G ' < 0.5 and G m ' ( k - 1 ) > 4
mg / dL / min , where ##EQU00001## .DELTA. G ' = G m ( k ) - G m (
k - 1 ) t ( k ) - t ( k - 1 ) - G F ' ( k - 2 ) G F ' ( k - 2 )
##EQU00001.2## and ##EQU00001.3## G m ' ( k - 1 ) = G m ( k - 1 ) -
G m ( k - 2 ) t ( k - 1 ) - t ( k - 2 ) . ##EQU00001.4##
G.sub.F is the filtered CGM data and G'.sub.F is the calculated
rate of change. The calculations of G.sub.F and G'.sub.F are
illustrated below in the data filtering and core algorithm
sections, respectively. If a shift is detected, the previous points
are shifted as follows:
G.sub.F(j)=G.sub.F(j)+residual(k-1)j=k-A-1:k-2
where A=number of subsequent alarms required to emit warning of
hypoglycemia and
residual(k-1)=G.sub.m(k-1)-[G'.sub.F(k-2).times.(t(k-1)-t(k-2))+G.sub.F(-
k-2)].
If a shift is detected, G.sub.F(k-1) is re-calculated using the
updated G.sub.F vector.
1.3 Data Filtering
[0035] Due to electrical noise and interference, the CGM data is
often noisy; hence filtering the data using physiologically-based
parameters allows the data to more accurately reflect the blood
glucose. The algorithm filters the data using a noise-spike filter
to reduce the impact of noise spikes, derived as follows:
G F , NS ( k ) = { G m ( k ) if G m ( k ) - G F ( k - 1 ) .ltoreq.
.DELTA. G G F ( k - 1 ) - .DELTA. G if ( G F ( k - 1 ) - G m ( k )
) > .DELTA. G G F ( k - 1 ) + .DELTA. G if ( G m ( k ) - G F ( k
- 1 ) ) > .DELTA. G , ##EQU00002##
where k is the sampling instant, G.sub.F(k-1) is the previous
filtered value, G.sub.F,NS(k) is the filtered value resulting from
the noise-spike filter, G.sub.m(k-1) is the measurement, and
.DELTA.G is the maximum allowable rate of change (Seborg, et al.,
2011, supra). The data are then passed through a low pass filter to
damp high frequency fluctuations from electrical noise, written as
follows:
G F ( k ) = .DELTA. t .tau. F + .DELTA. t G F , NS ( k ) + ( 1 -
.DELTA. t .tau. F + .DELTA. t ) G F ( k - 1 ) , ##EQU00003##
[0036] where .DELTA.t is the sampling time, .tau..sub.F is the
filter time constant, and G.sub.F is the filtered value (Seborg et
al., 2011, supra). A dimensionless parameter, .alpha., is defined
as follows:
.alpha. .DELTA. t .tau. F + .DELTA. t , ##EQU00004##
and varies from 0 to 1 (0 not included), with the filtered value
equaling the measurement if .alpha. equals 1, and the measurement
being ignored as a approaches 0.
1.4 Interpolation
[0037] Dropped measurements can lead to missing data points. In
order to allow the HMS to make a prediction even when points are
missing, these points will be interpolated in order to allow a
prediction to be made at that instance in time. The algorithm then
interpolates gaps of up to 20 minutes using linear
interpolation:
G F ( k - 1 / 2 ) = G F ( k - 1 ) + ( t ( k - 1 / 2 ) - t ( k - 1 )
) ( G F ( k ) - G F ( k - 1 ) ) ( t ( k ) - t ( k - 1 ) ) ,
##EQU00005##
where G.sub.F(k-1/2) is the interpolated point, halfway between
t(k-1) and t(k). Both the G.sub.F and t vectors are updated to
include the interpolated point.
TABLE-US-00001 TABLE 1-1 Explanation of symbols in Missed Point
Handling flow chart. Symbol Value Unit Interpretation A 1 -- Alarm
Requirement: # of subsequent positive flags for alarm Cmax 0.5 --
Maximum change: limits difference of G' before and after offset to
detect shift gap.sub.max 20 minutes Maximum gap to interpolate. If
this is exceeded, algorithm waits for enough points after the gap
to predict. gap.sub.min 7 minutes Minimum gap to interpolate
G'.sub.F -- mg/dL/min Estimated previous rate of change, used for
missing point extrapolation. G'.sub.m(k-1) -- mg/dL/min Slope of
previous two points, used for shift determination. G m ' ( k - 1 )
= G m ( k - 1 ) - G m ( k - 2 ) t ( k - 1 ) - t ( k - 2 )
##EQU00006## G'.sub.max 4 mg/dL/min Maximum rate of change for
shift detection low pass filter -- mg/dL G F ( k ) = .DELTA. t
.tau. F + .DELTA. t G F , NS ( k ) + ( 1 - .DELTA. t .tau. F +
.DELTA. t ) G F ( k - 1 ) ##EQU00007## noise spike filter -- mg/dL
G F , NS ( k ) = { G m ( k ) if G m ( k ) - G F ( k - 1 ) .ltoreq.
.DELTA. G G F ( k - 1 ) - .DELTA. G if ( G F ( k - 1 ) - G m ( k )
) > .DELTA. G G F ( k - 1 ) + .DELTA. G if ( G m ( k ) - G F ( k
- 1 ) ) > .DELTA. G ##EQU00008## PR 3 -- A - 1 + order of the G'
calculation residual -- mg/dL Used to update previous points when
shift is detected. residual (k - 1) = G.sub.m(k - 1) - [G'.sub.F(k
- 2) .times. (t(k - 1) - t (k - 2)) + G.sub.F(k - 2)] TT -- minutes
Last treatment time: used to determine it is too soon to alarm
after previous alarm .DELTA.G 3* .DELTA.t mg/dL Maximum allowable
rate of change for the noise spike filter. .DELTA.G' -- -- Used in
shift detection to detect offset with similar rates of change
before and after offset. .DELTA. G ' = G m ( k ) - G m ( k - 1 ) t
( k ) - t ( k - 1 ) - G F ' ( k - 2 ) G F ' ( k - 2 ) ##EQU00009##
.tau..sub.F 3 minutes Filter time constant. .DELTA.t 5 minutes
Sampling time: this may be longer if points are missing.
2. Core Algorithm
[0038] In the core algorithm, the rate of change is calculated to
make a prediction and issue an alarm if a hypoglycemic event is
imminent. The rate of change is calculated and the trajectory is
projected through the hypoglycemia threshold, TH, to decide if a
hypoglycemic event will occur within the prediction horizon, PH.
The rate of change calculation is as follows, using the first
derivative of the Lagrange interpolation polynomial (Dassau et al.,
2010, supra):
G F ' ( j ) .apprxeq. t ( j ) - t ( j - 1 ) ( t ( j - 2 ) - t ( j -
1 ) ) ( t ( j - 2 ) - t ( j ) ) G F ( j - 2 ) + t ( j ) - t ( j - 2
) ( t ( j - 1 ) - t ( j - 2 ) ) ( t ( j - 1 ) - t ( j ) ) G F ( j -
1 ) + 2 t ( j ) - t ( j - 2 ) - t ( j - 1 ) ( t ( j ) - t ( j - 1 )
) ( t ( j ) - t ( j - 2 ) ) G F ( j ) ##EQU00010##
where j=k-A+1:k. The following logic is then implemented: [0039] if
G.sub.F(k)<70 mg/dL and G'.sub.F(k)<-0.1 mg/dL/min, the Alarm
Mode is activated [0040] else, if G.sub.F(k)<110 mg/dL and -3
mg/dL/min<G'.sub.F(k)<0 mg/dL/min and
[0040] ( TH - G F ( j ) ) G F ' ( j ) < PH ##EQU00011##
.A-inverted. j = k - A + 1 : k , ##EQU00011.2##
the Alarm Mode is activated
[0041] A flow diagram of the module can be seen in FIG. 2 (Flow
chart of the Core Algorithm module, with terms detailed in Table
2-1.
TABLE-US-00002 TABLE 1-2 Explanation of symbols in Core Algorithm
module flow chart. Symbol Value Unit Interpretation A 1 -- Alarm
Requirement: # of subsequent positive flags for alarm
G'.sub.decrease -0.1 mg/dL/min Decreasing G': cutoff used when G is
below TH.sub.hypo to determine if the trend is negative. G'.sub.F
-- mg/dL/min Estimated current rate of change, using Lagrange
inter- polation polynomials. G F ' ( j ) .apprxeq. t ( j ) - t ( j
- 1 ) ( t ( j - 2 ) - t ( j - 1 ) ) ( t ( j - 2 ) - t ( j ) ) G F (
j - 2 ) - t ( j ) - t ( j - 2 ) ( t ( j - 1 ) - t ( j - 2 ) ) ( t (
j - 1 ) - t ( j ) ) G F ( j - 1 ) + 2 t ( j ) - t ( j - 2 ) - t ( j
- 1 ) ( t ( j ) - t ( j - 1 ) ) ( t ( j ) - t ( j - 2 ) ) G F ( j )
##EQU00012## G'.sub.maxdrop -3 mg/dL/min Maximum drop of G: cutoff
used for alarming. If drop is greater than this, it is considered
non-physiologic and will not alarm. G'.sub.mindrop -0.5 mg/dL/min
Minimum drop of G: cutoff used for alarming. If G is not dropping,
hypoglycemia is not imminent. PH 15 minutes Prediction Horizon:
time through which the prediction is projected. TH.sub.activation
110 mg/dL Activation threshold: alarm can only be triggered when G
is below this threshold to focus on danger of imminent
hypoglycemia. TH.sub.hypo 70 mg/dL Hypoglycemia threshold:
prediction is compared against this to determine danger of imminent
hypoglycemia. TTL -- minutes Projected time to crossing
TH.sub.hypo. TTL ( j ) = ( TH - G F ( j ) ) G F ' ( j )
##EQU00013##
Alarm Mode
[0042] The alarm mode will issue an audible and visible alarm and
activate E911, sending a short message service (SMS) to the
attending physician. If any alarms have been issued and
acknowledged in the past 30 minutes, no alarm is issued. If not, a
version of the following message will pop up for predicted
hypoglycemia, informing the clinicians of impending hypoglycemia,
the current rate of fall, and the time to predicted low; see, FIG.
3-1 for screenshot of the impending hypoglycemia pop-up window.
[0043] Data from user input (Accept or Ignore) will go to the
algorithm for use as a lockout window before subsequent alarms.
Should they accept, clinicians will then administer 16 g of rescue
carbohydrates to the subject. Also, if the threshold has been
crossed without alarms occurring and the CGM values continue to
fall, a version of the message of FIG. 3-2 (Representation of the
message when the CGM is below 70 mg/dL) will appear.
[0044] These figures will also be sent in a multimedia messaging
service (MMS) to the physician in charge. This adds redundancy to
ensure that treatment is given. A flow diagram of the module can be
seen in FIG. 3-2 with terms detailed in Table 3-1.
TABLE-US-00003 TABLE 3-1 Explanation of symbols in Alarm Mode
module flow chart. Symbol Value Unit Interpretation TH.sub.hypo 70
mg/dL Hypoglycemia threshold: prediction is compared against this
to determine danger of imminent hypoglycemia. TT -- minutes Last
treatment time: used to determine it is too soon to alarm after
previous alarm LT 30 minutes Treatment lockout time: minimum time
allowed between alarms
[0045] In a particular embodiment the disclosed Health Monitoring
System (HMS) is adapted for us in conjunction with an Artificial
Pancreas (AP) Device for type 1 diabetes (T1DM) patients using a
model-predictive control (MPC) algorithm (or MPC, PID, PD, FL,
NMPC, etc.) with a subcutaneous insulin delivery pump and a
subcutaneous continuous glucose monitor.
[0046] The AP device is composed of the Artificial Pancreas System
platform (APS.COPYRGT.) developed by the University of California,
Santa Barbara (UCSB) and Sansum Diabetes Research Institute (SDRI).
The APS.COPYRGT. is the current leading research platform used in
this arena. It has been safely used in over 100 individual clinical
sessions at eight leading clinical research centers around the
world. This AP device is a closed-loop insulin pump/continuous
glucose monitor (CGM) system regulated by a proprietary control
algorithm, and comprises:
[0047] (1) Artificial Pancreas System (APS.COPYRGT.) platform
(version 0.3.0) documented in MAF-1625 including the following
insulin pump and CGM:
[0048] OneTouch.RTM. Ping.RTM. Glucose Management System by Animas
Corporation (K080639 and MAF-1777). It is also called a Continuous
Subcutaneous Insulin Infusion (CSII) pump; and
[0049] DexCom.TM. SEVEN PLUS by DexCom.TM. Corporation (P050012 and
MAF-1564);
[0050] Interface to connect these components is programmed in
MATLAB.RTM. language (revision 2009b);
[0051] Accessory hardware to connect the components together;
[0052] (2) Control algorithms including the following components
(FIG. 4):
[0053] a zone Model Predictive Control (zone-MPC) algorithm that
automatically regulates the rate of insulin delivery based on the
glucose level of the patient, historical glucose measurement and
anticipated future glucose trends, and patient specific
information; and
[0054] a Health Monitoring System (HMS) algorithm that adds an
independent safety layer to the overall system. The HMS analyzes
CGM data and CGM trends in anticipation of impending hypoglycemia.
The HMS issues electronic, visual and/or audio alerts in response
to impending hypoglycemia (e.g. within 15 minutes), such as on the
AP device screen, with a request for the investigator to intervene
and treat the subject, e.g. with 16 g carbohydrate. A secondary
alert may be sent as a text message, such as to the clinical team,
that hypoglycemia is predicted and may also suggest taking outside
action, such as eating carbohydrates, in order to prevent
hypoglycemia. For example, the HMS will send a warning message when
predicting that glucose level by CGM will be <70 mg/dL in the
following 15 minutes, and the visual and audio alarms appear on the
AP device screen as shown by FIG. 3-1.
[0055] A secondary redundant alert is also sent via text graph to
the clinical team. The text can be received on any cell phone,
while the added graph (MMS) message with the chart can only be
received on "smart phones". The text and graphic messages indicate
that hypoglycemia is predicted within the next 15 minutes (or less)
and recommend taking outside intervention to prevent predicted
hypoglycemia and treat the subject with carbohydrates. The SMS and
MMS messages are redundant alerts to the audio and visual alerts on
the AP device screen. The visual pop-up window on the AP device
computer interface must be acknowledged. [0056] If the investigator
selects the "ignore" button of the HMS warning, at the next cycle,
i.e. 5 minutes later, if the prediction is that glucose
concentration is predicted to be <70 mg/dL in the following 15
minutes, a new alarm will sound and appear. [0057] If the
investigator selects the "accept" button and the subjects is
treated with carbohydrates as recommended, the system will perform
a new analysis at the next cycle, but the alarm will not be
activated for 30 minutes. If after 30 minutes the prediction is for
a risk of hypoglycemia (<70 mg/dL), then a new alarm will occur.
If the ingestion of carbohydrates prevented hypoglycemia, then, no
alarm will occur.
[0058] FIG. 5 show an example of the text message that is sent to
the clinician. The same text message can be sent to any cell phone,
and if the phone is a "smart phone", it will also receive the
trending visual plot of the glucose level and its prediction
trend.
[0059] Real-time prediction of pending adverse events by the Health
Monitoring System (HMS) allows prevention by either a corrective
action or shifting to manual control. This invention is based on
CGM data that provides a reliable layer of protection to insulin
therapy.
[0060] The first module in the HMS is a real-time hypoglycemia
prediction algorithm that includes a projection based on a short
term linear extrapolation of the glucose profile. This algorithm
first processes the data using a filter, interpolation of missed
points, and calibration detection. The risk of imminent
hypoglycemia is then calculated, and, if warranted, an audible and
visual alarm is sounded. In addition, the information about the
current state of the system and the prediction of hypoglycemia is
sent to the physician in charge via SMS and MMS. The mitigation of
this event is to consume 16 g of carbohydrate, which should
minimize the risk of severe hypoglycemia.
[0061] The systems and methods of the disclosure can be implemented
in a computer or processor operably-associated with continuous
glucose monitoring (CGM) devise and/or an insulin diabetes system
or pump. The HMS may incorporate a hypoglycemia prediction
algorithm (HPA) such as disclosed in U.S. Ser. No. 61/357,409,
filed Jun. 22, 2010, and the core algorithm may embody a numerical
logical algorithm that feeds a three-point calculated rate of
change using backward difference approximation and the current
glucose value into logical expressions to detect impending
hypoglycemia. The logical expressions verify that the rate of
change is both negative and within an acceptable range as well as
that the CGM glucose values are within predefined boundaries and
that a pending hypoglycemic event is predicted within the threshold
time window. Numerical logical algorithm provides insensitivity to
sensor signal dropouts and easy tuning.
[0062] In one aspect the invention effectively transforms CGM data
into a physicality that is an audio and/or visual alert that
hypoglycemia is imminent. In another aspect the invention
effectively transforms CGM data into a negative feedback signal and
send it to an insulin delivery device, which consequently actuates
the delivery device, such as by restricting fluid flow, adjusting a
fluid valve, reducing or shutting off a pump, etc.
[0063] The foregoing examples and detailed description are offered
by way of illustration and not by way of limitation. All
publications and patent applications cited in this specification
are herein incorporated by reference as if each individual
publication or patent application were specifically and
individually indicated to be incorporated by reference. Although
the foregoing invention has been described in some detail by way of
illustration and example for purposes of clarity of understanding,
it will be readily apparent to those of ordinary skill in the art
in light of the teachings of this invention that certain changes
and modifications may be made thereto without departing from the
spirit or scope of the appended claims
* * * * *