Early recognition of change to pathophysiologic state of dysglycemia
Conditions of pathophysiologic dysglycemia, associated with type 1 diabetes, can be detected early, including in children. Glycemia data is collected by continuous glucose monitoring devices connected to a cloud server. The server learns person-specific patterns of glycemia and trains a model accordingly, which may serve as a baseline for comparison or as the basis for prediction of future patterns. Such models are conditioned by risk factors and various concurrent activities such as exercise and consumption of sugars. Person-specific models trained at different time periods give different simulated patterns of glycemia. Deviations between the simulated patterns, or between the predicted and actual patterns, can indicate progression of pathophysiologic dysglycemia and type 1 diabetes. These models and comparisons can be further interpreted by the software to result in a reported recommendation for additional diagnostic testing or a reported conclusion of diagnosis and recommendation of preventative intervention.
1 . A computer-implemented method of detecting early stage diabetes to identify an opportunity for early preventative disease-modifying therapy, the method comprising:
during a first time period, the first time period being long enough for at least several occasions between states of euglycemia and dysglycemia:
configuring a glucose sensor to sample first glucose monitoring data for a person at a first sampling frequency equal to or higher than a transition frequency of transitions between the state of euglycemia and the state of dysglycemia, the person being antibody positive (Aby+);
receiving at a remote server the first glucose monitoring data;
configuring a second sensor to obtain first additional person-specific data for the person at least at the first sampling frequency, the first additional person-specific data including one or more types of additional biometric data or activity data of the person;
receiving at the remote server the first additional person-specific data;
receiving at the remote server first food data including first foods eaten during the first time period and first associated times of day;
during each of one or more second time periods, the one or more second time periods initiating a duration after the first time period, each of the one or more second time periods being long enough for at least several occasions between the states of euglycemia and dysglycemia:
configuring the glucose sensor to sample second glucose monitoring data for the person at a second sampling frequency equal to or higher than the transition frequency of transitions between the state of euglycemia and the state of dysglycemia;
receiving at the remote server the second glucose monitoring data;
configuring the second sensor to obtain second additional person-specific data for the person at least at the first sampling frequency, the second additional person-specific data including the one or more types of additional biometric data or activity data of the person;
receiving at the remote server the second additional person-specific data;
receiving at the remote server second food data including second foods eaten during the second time period and second associated times of day;
using statistical analysis to generate a first person-specific model of first patterns in the first glucose monitoring data in relation to the first food data and the first additional person-specific data for the first time period;
using statistical analysis to generate in real time respectively one or more second person-specific models of second patterns in the second glucose monitoring data in relation to the second food data and the second additional person-specific data for the one or more second time periods;
using third food data, third additional person-specific data and the first person-specific model to generate a first simulated glucose pattern for the person based on the first time period;
using the third food data, the third additional person-specific data and the second person-specific model to generate in real time a second simulated pattern for the person based on at least one of the one or more second time periods;
computing in real time glucose-pattern deviation information between the first simulated pattern and the at least one of the one or more second simulated patterns for the person;
monitoring in real time the glucose-pattern deviation information for an actual diabetes pathophysiologic change condition in the person indicative that early preventative disease-modifying therapy may be warranted, the actual diabetes pathophysiologic change condition generated from a machine learned model trained on glycemic data gathered from at least a set of individuals in the population having similar characteristics as the person; and
when the actual pathophysiologic change condition based on the glucose pattern deviation information for the person is satisfied, generating a notification in real time to at least one of the person, a caregiver or a healthcare provider that early preventative disease-modifying therapy may be warranted.
2 . The method of claim 1 wherein the notification includes a probability of a positive result from an oral glucose tolerance test, the probability computed using a population-wide model of oral glucose tolerance test results relative to computed pattern deviation information.
3 . The method of claim 1 wherein the notification includes at least one of a diagnostic recommendation or an intervention recommendation to a remote electronic health record database via a computer network.
4 . The method of claim 1 wherein computing the pattern deviation information comprises at least one of:
(a) computing a difference in amplitude; or
(b) computing a difference in rate of change.
5 . The method of claim 3 further comprising obtaining an antibody test result for the person, wherein at least one of the diagnostic recommendation or the intervention recommendation is computed using a population-wide model as a function of the antibody test result.
6 . The method of claim 3 wherein at least one of the diagnostic recommendation or the intervention recommendation is one or more of:
that additional diagnostic testing is warranted;
that a diagnosis of disease progression is warranted;
that a diagnosis of Type I Diabetes, Stage 2 should be considered; or
that intervention with the early preventative disease-modifying therapy may be warranted.
7 . The method of claim 3 further comprising obtaining at least one of a genetic test result or a history of intervention with disease-modifying treatment for the person, wherein at least one of the diagnostic recommendation or the intervention recommendation is computed as a function of at least one of:
the genetic test result; or
the history of intervention with disease-modifying treatment.
8 . The method of claim 1 performed by the remote server in response to a request through a web API.
9 . The method of claim 1 wherein the glucose sensor and the second sensor are parts within a single device.
10 . The method of claim 1 wherein computing the pattern deviation information comprises computing a difference in a parameter learned from the first and/or second additional patient-specific data.
11 . The method of claim 1 wherein the duration is about a month.
12 . The method of claim 1 wherein the first person-specific model is generated from a pre-trained population-wide model to reduce computations needed to generate the first person-specific model.
13 . A non-transitory computer readable medium storing code that causes a computer to:
during a first time period, the first time period being long enough for at least several occasions between states of euglycemia and dysglycemia:
configure a glucose sensor to sample first glucose monitoring data for a person at a first sampling frequency equal to or higher than a transition frequency of transitions between the state of euglycemia and the state of dysglycemia, the person being antibody positive (Aby+);
receive at a remote server the first glucose monitoring data;
configure a second sensor to obtain first additional person-specific data for the person at least at the first sampling frequency, the first additional person-specific data including one or more types of additional biometric data or activity data of the person;
receive at the remote server the first additional person-specific data;
receive at the remote server first food data including first foods eaten during the first time period and first associated times of day;
during each of one or more second time periods, the one or more second time periods initiating a duration after the first time period, each of the one or more second time periods being long enough for at least several occasions between the states of euglycemia and dysglycemia:
configure the first glucose sensor to sample second glucose monitoring data for the person at a second sampling frequency equal to or higher than the transition frequency of transitions between the state of euglycemia and the state of dysglycemia;
receive at the remote server the second glucose monitoring data;
configure the second sensor to obtain second additional person-specific data for the person at least at the first sampling frequency, the second additional person-specific data including the one or more types of additional biometric data or activity data of the person;
receive at the remote server the second additional person-specific data;
receive at the remote server second food data including second foods eaten during the second time period and second associated times of day;
use statistical analysis to generate a first person-specific model of first patterns in the first glucose monitoring data in relation to the first food data and the first additional person-specific data for the first time period;
use statistical analysis to generate in real time respectively one or more second person-specific models of second patterns in the second glucose monitoring data in relation to the second food data and the second additional person-specific data for the one or more second time periods;
use third food data, third additional person-specific data and the first person-specific model to generate a first simulated glucose pattern for the person based on the first time period;
use the third food data, the third additional person-specific data and the second person-specific model to generate in real time a second simulated pattern for the person based on at least one of the one or more second time periods in real time;
compute in real time glucose-pattern deviation information between the first simulated pattern and the at least one of the one or more second simulated patterns for the person;
monitor in real time the glucose-pattern deviation information for an actual diabetes pathophysiologic change condition in the person indicative that early preventative disease-modifying therapy may be warranted, the actual diabetes pathophysiologic change condition generated from a machine learned model trained on glycemic data gathered from at least a set of individuals in the population having similar characteristics as the person; and
when the actual pathophysiologic change condition based on the glucose pattern deviation information for the person is satisfied, generate a notification in real time to at least one of the person, a caregiver or a healthcare provider that early preventative disease-modifying therapy may be warranted.
14 . The computer readable medium of claim 13 wherein the notification includes a probability of a positive result from an oral glucose tolerance test, the probability computed using a population-wide model of oral glucose tolerance test results relative to computed pattern deviation information.
15 . The computer readable medium of claim 13 wherein the notification includes at least one of a diagnostic recommendation or an intervention recommendation to a remote electronic health record database via a computer network.
16 . The computer readable medium of claim 13 wherein the pattern deviation information is computed by one of
(a) computing a difference in amplitude; or
(b) computing a difference in rate of change.
17 . The computer readable medium of claim 15 that further causes the computer to obtain at least one of a history of prior intervention with disease-modifying treatment or an antibody test result for the person, wherein at least one of the diagnostic recommendation or the intervention recommendation is computed as a function of at least one of:
the antibody test result; or
the history of prior intervention with disease-modifying treatment.
18 . The computer readable medium of claim 15 wherein at least one of the diagnostic recommendation or the intervention recommendation is one or more of:
that additional diagnostic testing is warranted;
that a diagnosis of disease progression is warranted;
that a diagnosis of Type I Diabetes, Stage 2 should be considered; or
that intervention with a preventative disease-modifying therapy may be warranted.
19 . The computer readable medium of claim 15 that further causes the computer to obtain at least one of:
a genetic test result; or
a history of intervention with disease-modifying treatment for the person, wherein at least one of the diagnostic recommendation and the intervention recommendation is computed as a function of at least one of:
the genetic test result; or
the history of intervention with disease-modifying treatment.
20 . The computer readable medium of claim 13 wherein the obtaining is performed by the remote server in response to a request through a web API.
21 . The computer readable medium of claim 13 wherein the glucose sensor and the second sensor are parts within a single device.
22 . The computer readable medium of claim 13 wherein computing the pattern deviation information comprises computing a difference in a parameter learned from the first and/or second additional patient-specific data.
23 . The computer readable medium of claim 13 wherein the duration is about a month.
24 . The computer readable medium of claim 13 wherein the first person-specific model is generated from a pre-trained population-wide model to reduce computations needed to generate the first person-specific model.