Long term HbA1c prediction
Provided are methods of predicting HbA1c values.
1 . A method for predicting HbA1c values comprising:
collecting HbA1c time series data of a patient across a plurality of days at a plurality of times for each day of the plurality of days;
converting the HbA1c time series data into a histogram image formatted to indicate a distribution of numerical values in HbA1c time series data organized by a number of days in the plurality of days as a first coordinate in the histogram image and a bin into which a given numerical value of the numerical values in the HbA1c time series data is classified as a second coordinate in the histogram image;
analyzing the image using a convolutional neural network;
predicting a predicted HbA1c level for the patient at a future time point; and
in response to the predicted HbA1c level for the patient being greater than or equal to a predefined threshold at the future time point, prescribing a treatment plan for the patient for administration of a therapeutic agent before the future time point.
2 . The method of claim 1 , wherein the predefined threshold is 6.5.
3 . The method of claim 1 , wherein the therapeutic agent administered according to the treatment plan comprises metformin.
4 . The method of claim 1 , wherein the therapeutic agent administered according to the treatment plan comprises insulin.
5 . A method for evaluating diabetes progression comprising:
collecting HbA1c time series data of a patient across a plurality of days at a plurality of times for each day of the plurality of days;
converting the HbA1c time series data into an image that represents the HbA1c time series data organized using timing of when individual values in the HbA1c time series data as a first coordinate in the image;
analyzing the image using a convolutional neural network;
predicting a predicted HbA1c level for the patient at a future time point; and
in response to the predicted HbA1c level for the patient being greater than or equal to a predefined threshold at the future time point, prescribing a treatment plan for the patient for administration of a therapeutic agent before the future time point.
6 . The method of claim 5 , wherein the future time point is at least one week in the future from predicting the predicted HbA1c level.
7 . The method of claim 5 , wherein the future time point is at least two weeks in the future from predicting the predicted HbA1c level.
8 . The method of claim 5 , wherein the future time point is at least three weeks in the future from predicting the predicted HbA1c level.
9 . The method of claim 5 , wherein the future time point is at least four weeks in the future from predicting the predicted HbA1c level.
10 . The method of claim 5 , wherein the image is a binary image formatted to indicate a count of measurements in the HbA1c time series data that are organized by which day in the plurality of days the measurements in the HbA1c time series data were collected on as the first coordinate in the binary image and which range of a plurality of a ranges of values that a given measurement of the measurements in the HbA1c time series data is categorized into as a second coordinate in the binary image.
11 . The method of claim 5 , wherein the image is a histogram image, formatted to indicate a distribution of numerical values in the HbA1c time series data organized by a number of days in the plurality of days as the first coordinate in the histogram image and a bin into which a given numerical value of the numerical values in the HbA1c time series data is classified as a second coordinate in the histogram image.
12 . The method of claim 5 , wherein every address in the image according to pairings of the first coordinate and a second coordinate is populated with a color value based on the HbA1c time series data as organized in the image.
13 . A method, comprising:
collecting HbA1c time series data of a patient across a plurality of days at a plurality of times for each day of the plurality of days;
converting the HbA1c time series data into a binary image formatted to indicate a count of measurements in the HbA1c time series data that are organized by which day in the plurality of days the measurements in the HbA1c time series data were collected on as a first coordinate in the binary image and which range of a plurality of a ranges of values that a given measurement of the measurements in the HbA1c time series data are categorized into as a second coordinate in the binary image;
analyzing the binary image using a convolutional neural network;
predicting a predicted HbA1c level for the patient at a future time point at least one week in the future; and
in response to the predicted HbA1c level for the patient being greater than a predefined threshold, prescribing a treatment plan for the patient of at least one of metformin administration and insulin administration before the future time point.
14 . The method of claim 13 , wherein the plurality of days consists of fourteen days of continuous blood glucose (CGM) measurements from the patient.
15 . The method of claim 13 , wherein the HbA1c time series data is collected at fifteen minute intervals from the patient.