WO2018229209A1 - Algorithme de dosage d'insuline sur la base d'un profil de patient - Google Patents

Algorithme de dosage d'insuline sur la base d'un profil de patient Download PDF

Info

Publication number
WO2018229209A1
WO2018229209A1 PCT/EP2018/065845 EP2018065845W WO2018229209A1 WO 2018229209 A1 WO2018229209 A1 WO 2018229209A1 EP 2018065845 W EP2018065845 W EP 2018065845W WO 2018229209 A1 WO2018229209 A1 WO 2018229209A1
Authority
WO
WIPO (PCT)
Prior art keywords
algorithm
dose size
insulin
data
subject
Prior art date
Application number
PCT/EP2018/065845
Other languages
English (en)
Inventor
Oleksandr SHVETS
Tinna Björk ARADÓTTIR
Thomas Dedenroth Miller
Anuar IMANBAYEV
Original Assignee
Novo Nordisk A/S
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Novo Nordisk A/S filed Critical Novo Nordisk A/S
Priority to JP2019568745A priority Critical patent/JP7181900B2/ja
Priority to US16/622,384 priority patent/US20200227170A1/en
Priority to EP18729473.1A priority patent/EP3639168B1/fr
Priority to CN201880039363.2A priority patent/CN110753967A/zh
Publication of WO2018229209A1 publication Critical patent/WO2018229209A1/fr

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
    • G16H20/10—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to drugs or medications, e.g. for ensuring correct administration to patients
    • G16H20/17—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to drugs or medications, e.g. for ensuring correct administration to patients delivered via infusion or injection
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
    • G16H40/60—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices
    • G16H40/67—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices for remote operation

Definitions

  • the present disclosure relates generally to a device, system and method for assisting patients in managing pen based insulin treatment for diabetes, in which insulin injection doses are adapted to predicted risk of hypoglycemia (or hyperglycemia), the prediction being based on learning by machine learning the individuality of a patient and her/his profile as to historic blood glucose levels and injected insulin doses.
  • BG blood glucose
  • Insulin is a hormone that binds to insulin receptors to lower BG level by facilitating cellular uptake of glucose, amino acids, and fatty acids into skeletal muscle and fat and by inhibiting the output of glucose from the liver.
  • physiologic insulin secretions maintain normoglycemia, which affects fasting plasma glucose and postprandial plasma glucose concentrations.
  • the insulin secretion is impaired in Type 2 diabetes and early post-meal response is absent.
  • Type 1 diabetes there is very low to no insulin secretion at all.
  • Patients are usually provided with insulin medicament treatment regimens and the goal of these regimens is to maintain a desired fasting or prandial BG target level.
  • insulin medicament treatment regimens are based on titration algorithms that provide dose recommendations to the patients, the algorithms being based on retrospective use of historical blood glucose data, either self-monitored (SMPG) or Continuous Glucose Measuring (CGM) data, to calculate the dose recommendations.
  • SMPG self-monitored
  • CGM Continuous Glucose Measuring
  • Titration algorithms are often based on ADA (American Diabetes Association) Standards of care
  • titration algorithm is a 2-0-2 titration algorithm that recommends the patient to increase the dose by 2 units, if the fasting blood glucose level is too high, or continue with the same dose dosage as previous, if the fasting blood glucose level is at normal level, or decrease the dose size by 2 units, if the fasting blood glucose level is too low.
  • titration algorithms result normally in a reasonably good result however the dose recommendations are only based on the previous BG measurement data, i.e. acts only retrospectively and cannot predict future events.
  • a standard titration algorithm may give a good result with few BG fluctuations, but that's often not the case.
  • Patients often have changes in normal routine on selected days during the week. There are many different examples for that, e.g. they have a proper meal with low carbohydrates level or they are doing sports after the meal.
  • BG level may lead to changes in BG level, if the patient doesn't adjust bolus injections, which may lead to risk of hypoglycemic events.
  • the BG level will fluctuate up and down every time an adjustment after a hypoglycemic event is made. Furthermore, the patient has to recover from the hypoglycemic event, which is very unfortunate.
  • hypoglycemic events may result in unnecessary fluctuations in BG levels and may lead to worse adherence and long term complications for the patients, which is very unfortunate. Besides this many patients usually don't track these events and thus don't ask for advices from health care professionals (HCP). Due to this it is important to help by calculating all possible risks associated with a given advice for bolus as well as for basal insulin injections.
  • HCP health care professionals
  • CGM chronic myetitrating
  • Post prandial blood glucose is not measured at all.
  • titration systems/methods that is able to predict a risk of hypoglycemic event and provide an insulin dose that will keep the patient at the normoglycemia level. It will reduce the nos. of hypoglycemic events significantly and result a less fluctuations in BG level.
  • US 2015/0190098 disclose an adaptive advisory control interactive process involving algorithm based assessment and communication of physiologic and behavioural parameters and patterns assisting diabetics optimising their glycemic control. It provides an algorithm to retroactively compute a risk-based insulation attenuation factor to the subject's record of insulin delivery.
  • US 2014/0350369 discloses a system and method that provides a glucose report for determining glycemic risk based on an ambulatory glucose profile of glucose data over a time period, a glucose control assessment based on median and variability of glucose, and indicators of high glucose variability. It uses statistical methods to provide titration guidance such that glucose targets are reached in less time, with less likelihood of hypoglycaemia.
  • CPHS CGM-Based Prevention of hypoglycaemia System
  • US 6,923,763 and US 2017/0068790 also disclose systems and methods for predicting hypoglycemia.
  • WO2011/157402 discloses methods and systems for diabetic patients to optimize their administered insulin dosage.
  • None of the above references addresses the problem of providing an insulin titration system for patients being on pen based injections that can predict hypoglycemic events and automatically adjust a first calculated insulin dose size that would result in a hypoglycemic event down to a lower second dose size in order to prevent a hypoglycemic event and communicating this to the user, this happening without any interaction with the user.
  • the present disclosures addresses the need in the art for devices, systems and methods for providing improved insulin titration algorithm for pen based insulin treatment and resulting in fewer fluctuations in the patients BG levels and minimizes the risk of a hypoglycemic events. More particular, the invention provides a device, system and method improving insulin titration by adding a risk predictor module to a standard basal and bolus titration algorithms.
  • the risk predictor can predict a risk of hypoglycemia based on learning the patient's individual treatment pattern and thereby take into account future periodic deviations, where the blood glucose level is low (or high), and provide a corrected insulin dose to prevent a hypoglycemic event.
  • a first aspect of the present disclosure provides device for predicting risk of hypoglycemia and adjusting an insulin medicament dose size for a subject accordingly to prevent a predicted hypoglycemic event.
  • the device comprises one or more processors and a memory, the memory storing a data structure comprising a prescribed insulin medicament dosage regimen that specifies a first rule based titration algorithm for computing and recommending a first dose size as an adjustment to a current dosage regimen based upon a first data set.
  • the first data set comprises a plurality of blood glucose (BG) measurements taken over a time course, and for each respective measurement in the plurality of measurements, a corresponding timestamp representing when in the time course the respective measurement was made.
  • BG blood glucose
  • the first data set further comprises a plurality of injected insulin dose size taken over a time course, and for each respective injected dose size in the plurality of injected doses, a corresponding timestamp representing when in the time course the respective dose was injected and a respective typestamp of the insulin medicament injected.
  • the first data set comprises also a minimum target fasting and post prandial blood glucose level (mg/dl) and a maximum target for fasting and post prandial blood glucose level (mg/dl).
  • the data structure further comprises a risk prediction module adapted to predict risk of hypoglycemia and specifying the following machine learning algorithm modules; i) a short term risk prediction classification machine learning algorithm using a Gaussian Kernel and regularization to compute, based upon the last up to 3 hours of BG measurements from said first data set, the probability of hypoglycaemia occurring within a time period of up to 3 hours from start of computing the risk, ii) a long term risk prediction classification machine learning algorithm using a Gaussian Kernel and regularization to compute, based upon the last up to 24 hours of BG data from said first data set, the probability of hypoglycaemia occurring within a time period of up to 24 hours from start of computing the risk, iii) a subject's pattern recognition machine learning algorithm to analyse correlations between said BG measurements and insulin dose size data of said first data set and identify historic hyper- and hypoglycemic events over a predetermined period of time and the severity as to blood glucose level of each of these events.
  • a risk prediction module
  • the memory further stores instructions that, when executed by the one or more processors, perform a method comprising : responsive to a request from the subject to compute a recommended dose for a basal or bolus insulin; a) updating the first data set with latest available data, b) computing a post prandial blood glucose level (PPGL) (for bolus insulin dose request) as the lowest weighted average of at least the last three days post prandial blood glucose values in the first set of data, c) computing a titration blood glucose level (TGL) (for basal insulin dose
  • ⁇ a first dose size being decreased compared to the current dosage regimen for the subject, if said TGL or PPGL is below the said minimum targets for fasting or post prandial blood glucose levels, or
  • ⁇ a first dose size being equal to the current dosage regimen for the subject if said average TGL or PPGL is within the range between said the minimum and maximum targets for fasting or post prandial blood glucose levels, e) employing said short term risk prediction machine learning classification
  • the step of employing said short term risk prediction machine learning classification algorithm comprises preparing from the first data set specific data sets of historic BG measurements obtained within specific time intervals of the last 15 to 180 minutes, and running each prepared data set through a support vector machine classifier with Gaussian Kernel and regularization to compute said output x ⁇ indicating risk of hypoglycemia within a future time horizon of between 15 to 180 minutes, respectively.
  • Said specific time intervals can be any time intervals within the 15 to 180 minutes, such as the last 15 minutes, 17 minutes, 20 minutes, 30 minutes, 35 minutes, 45 minutes, 60 minutes or more, such as 120 minutes 135 minutes or more, and the future time horizon can be any future time horizon such as 15 minutes, 17 minutes, 20 minutes, 30 minutes, 35 minutes, 45 minutes, 60 minutes or more, such as 120 minutes 135 minutes or more.
  • the step of employing said long term risk prediction machine learning classification algorithm comprises preparing from the first data set data sets of historic BG
  • Said specific time intervals for the long term risk prediction algorithm can be any time intervals within the last 12 to 24 hours, such as the last 12.5 hours, 14 hours, 15 hours, 16 hours, 17 hours, 18.5 hours, 21 hours or more, and the future time horizon can also be any future time horizon such as 12 to 24 hours, such as the last 12.5 hours, 14 hours, 15 hours, 16 hours, 17 hours, 18.5 hours, 21 hours or more, up to 24 hours.
  • the step of employing said subject's pattern recognition algorithm comprises preparing from the first data set data sets of historic BG measurements obtained within specific time intervals of the last 15-180 minutes, and running each prepared data set through an single class support vector machine classifier algorithm to analyse correlations between said BG measurements of said data sets and corresponding insulin dose size data of said data sets and identify hyper- and hypoglycemic events within said time interval and the severity as to blood glucose level of each of these events.
  • Said specific time intervals can be any intervals within the 15 to 180 minutes, such as the last 15 minutes, 17 minutes, 20 minutes, 30 minutes, 35 minutes, 45 minutes, 60 minutes or more, such as 120 minutes 135 minutes or more.
  • the device provides an improved titration algorithm, where the risk prediction module is able to predict if a dose size of insulin, as recommended according the first rule based titration algorithm, will result in hypoglycemic (or hyperglycemic) event, calculate a safer second dose size and adjust the first dose size down to the second safer dose size.
  • the risk prediction module is able to predict if a dose size of insulin, as recommended according the first rule based titration algorithm, will result in hypoglycemic (or hyperglycemic) event, calculate a safer second dose size and adjust the first dose size down to the second safer dose size.
  • the first rule based titration algorithm can be any titration algorithm suitable for the given treatment regimen for the patient. Examples of algorithms are shown in the below tables:
  • the risk prediction module may either calculate a complete new second dose size or it may optionally calculate a correction value to the first dose size.
  • the output from the risk prediction module could either be the 15 units dose size as a new dose or be a correction value of 5 units to the computed first dose of 20 units.
  • the prescribed insulin medicament regimen includes a long acting basal insulin medicament dosage regimen, wherein the first rule based titration algorithm is adapted to compute a first dose size for the basal insulin medicament dosage and the risk prediction module is adapted compute a second dose size for the basal insulin resulting in a non-hypoglycemia state for the subject.
  • Said dosage regimen may consist of a single insulin medicament having duration of action that is between 12 and 24 hours or a mixture of insulin medicaments that collectively have duration of action that is between 12 and 24 hours. Examples of such long acting insulin
  • the prescribed insulin regimen also includes a short acting bolus insulin medicament dosage regimen, wherein the first algorithm is adapted to compute a first dose size for the bolus insulin medicament dosage and the risk prediction module is adapted to compute a second dose size for the bolus insulin resulting in a non- hypoglycemia state for the subject.
  • Said dosage regimen may comprise a single insulin medicament having a duration of action that is between three to eight hours or a mixture of insulin medicaments that collectively have a duration of action that is between three to eight hours.
  • Examples of such short acting insulin medicaments include, but are not limited, to Lispro (HUMALOG, insulin lispro [rDNA origin] injection, Eli Lilly and Company) and Aspart (NOVOLOG, insulin aspart [rDNA origin] injection, NOVO NORDISK Inc).
  • the short term risk prediction algorithm may use at least the previous three days of historic BG data from said first data set, but it could use more, such as the last four, five or even seven days or more. The more data available the more precise the prediction is, but the data used may be less than the previous three days, e.g. only the last two days or last one day or even less, such as last 12 hours or last 6 hours or last 4 hours.
  • the machine learning model is preferably a support vector machine classifier, but
  • the long term risk prediction algorithm may use up to the previous three days of historic from said first data set, however it could be more such as previous four, five, six, seven or more days. The more data available the more precise the prediction is, but the data used may be less than previous three days, e.g. only the last two days. It is preferably based upon support vector machine classifier, but alternatively it could be based on RF (Random Forest) and/or K-NN (K-Nearest Neighbors methodology) or another suitable methodology.
  • RF Random Forest
  • K-NN K-Nearest Neighbors methodology
  • the machine learning methodologies is preferably based on classification but could also be based on regression analysis depending on what fits best for the short term and long term risk prediction.
  • the first data set may comprise data about all previous first doses computed by the first algorithm, second doses computed by the risk prediction module and timestamps for the time of requesting dose recommendations.
  • a second data set may be obtained comprising sets of historic data containing one or more of insulin sensitivity factor (ISF) estimations, Insulin on board (IoB) estimations, carbohydrates on board (CoB) estimations, and for each of the respective measurements a corresponding timestamp representing when in the time course the respective measurement.
  • the method of calculating a second dose size consists of a training stage and a prediction stage. During the training stage coefficients ( ⁇ ,) are identified indicating how much influence each the individual algorithm modules of the risk prediction module impacts a dose reduction. These coefficients will be used to calculate a second dose size.
  • the steps to identify coefficients (weights) include defining an effect factor Z being indicative of the effect the injected insulin doses from the first data set had on the blood glucose level of the subject.
  • Z depends on the target set for the various BG levels and an example of such targets is shown in the table below.
  • These targets are usually set by the patient's HCP and are included in the first set of data (minimum target fasting BG level (FGL2), minimum post prandial BG level (PPL2), a maximum target fasting BG level (FGH), a maximum post prandial BG level (PPH), and a hypoglycemic blood glucose alert level (FGL1/PPL1).
  • PPBG post prandial glucose level
  • FBG fasting blood glucose level
  • the dose should be lowered by 20% to prevent a hypoglycemic event, if the BG level is staying lower than PPL1 or FGL1.
  • the method then continues with defining actual number of units said first dose size should be lowered to prevent a predicted hypoglycemic event, the actual number being calculated based on minimization model (squared error function) : m
  • X is the input from the subject's pattern recognition algorithm, short term risk prediction algorithm and long term risk prediction algorithm
  • y is the safe dose sizes
  • m is number of individual insulin injections.
  • the set of values is used in a prediction stage to calculate the second dose size using the formula (1) above.
  • the device preferably comprises a wireless receiver, and wherein the blood glucose measurements of the first data set is obtained wirelessly from a CGM affixed to the subject and the insulin dose data of the first data set is obtained wirelessly from one or more insulin pens adapted to communicate with the wireless receiver.
  • the pen may either be a durable or a disposable injection pen.
  • Self-monitored blood glucose (SMPG) measurements finger prick
  • SMPG measurements could theoretically also be used but practically SMPG measurements are most often not taken frequently enough by the patients (usually only 1 - 2 times per day), which doesn't give sufficient data for a proper prediction calculation.
  • the successive measurements in the plurality of glucose measurements may be taken autonomously by the CGM device from the subject at an interval rate of 15 minutes or less, 3 minutes or less, or 1 minute or less.
  • the data structure further comprises a data set of historic data about the subject's historic geographical positions, cardiovascular activity, food intake, calendar events, heart rate, skin temperature, skin impedance and/or respiration, for each of the respective measurements a corresponding timestamp representing when in the time course the respective measurement was made.
  • This data set being used by the risk prediction module to predict a risk of hypoglycaemia.
  • a second aspect of the invention provides a system for assisting a subject in treating diabetes and comprising a device according to the first aspect integrated into an application running on a mobile device, such as a mobile phone, laptop or tablet, said application being adapted to at least communicate on said mobile device the first and second dose size recommendations to the subject.
  • a mobile device such as a mobile phone, laptop or tablet
  • the application being adapted to at least communicate on said mobile device the first and second dose size recommendations to the subject.
  • the BG data and insulin injection data for the first data set is wirelessly communicated directly from the CGM and injection pen(s) to the mobile device.
  • the mobile device could be a smart phone or a tablet.
  • a third aspect of the invention provides method for assisting a subject in treating diabetes via a computer system comprising one or more processors and a memory storing a data structure comprising; a prescribed insulin medicament dosage regimen that specifies a first rule based titration algorithm for computing and recommending a first dose size as an adjustment to a current dosage regimen based upon a first data set comprising :
  • BG blood glucose
  • a risk prediction module comprising the following machine learning algorithm modules: a short term risk prediction classification machine learning algorithm using an activation function to compute, based upon the last up to 3 hours of BG measurements from said first data set, the probability of hypoglycaemia occurring within a time period of up to 3 hours from start of computing the risk, a long term risk prediction classification machine learning algorithm using an activation function to compute, based upon the last up to 24 hours of BG data from said first data set, the probability of hypoglycaemia occurring within a time period of up to 24 hours from start of computing the risk, a subject's pattern recognition machine learning algorithm to analyse correlations between said BG measurements and insulin dose size data of said first data set and identify historic hyper- and hypoglycemic events over a predetermined period of time and the severity as to blood glucose level of each of these events, wherein the memory further stores instructions that, when executed by the on or more processors, perform a method comprising : responsive to a request from the subject to compute
  • ⁇ a first dose size being decreased compared to the current dosage regimen for the subject, if said TGL or PPGL is below the said minimum targets for fasting or post prandial blood glucose levels, or
  • ⁇ a first dose size being equal to the current dosage regimen for the subject if said average TGL or PPGL is within the range between said the minimum and maximum targets for fasting or post prandial blood glucose levels, e) employing said short term risk prediction machine learning classification
  • the step of employing said short term risk prediction machine learning classification algorithm comprises preparing from the first data set specific data sets of historic BG measurements obtained within specific time intervals of the last 15 to 180 minutes, and running each prepared data set through a support vector machine classifier with Gaussian Kernel and regularization to compute said output indicating risk of hypoglycemia within a future time horizon of between 15 to 180 minutes, respectively.
  • Said specific time intervals can be any time intervals within the 15 to 180 minutes, such as the last 15 minutes, 17 minutes, 20 minutes, 30 minutes, 35 minutes, 45 minutes, 60 minutes or more, such as 120 minutes 135 minutes or more, and the future time horizon can be any future time horizon such as 15 minutes, 17 minutes, 20 minutes, 30 minutes, 35 minutes, 45 minutes, 60 minutes or more, such as 120 minutes 135 minutes or more.
  • the step of employing said long term risk prediction machine learning classification algorithm comprises preparing from the first data set data sets of historic BG
  • Said specific time intervals for the long term risk prediction algorithm can be any time intervals within the 12 to 24 hours, such as the last 12.5 hours, 14 hours, 15 hours, 16 hours, 17 hours, 18.5 hours, 21 hours or more
  • the future time horizon can also be any future time horizon such as 12 to 24 hours, such as the last 12.5 hours, 14 hours, 15 hours, 16 hours, 17 hours, 18.5 hours, 21 hours or more, up to 24 hours.
  • the step of employing said subject's pattern recognition algorithm comprises preparing from the first data set data sets of historic BG measurements obtained within specific time intervals of the last 15-180 minutes, and running each prepared data set through an single class SVM classifier algorithm to analyse correlations between said BG measurements of said data sets and corresponding insulin dose size data of said data sets and identify hyper- and hypoglycemic events within said time interval and the severity as to blood glucose level of each of these events.
  • Said specific time intervals can be any time intervals within the 15 to 180 minutes, such as the last 15 minutes, 17 minutes, 20 minutes, 30 minutes, 35 minutes, 45 minutes, 60 minutes or more, such as 120 minutes 135 minutes or more.
  • each patient is risk-stratified in real time based on each patient's first set of data (also called "Patient's metric").
  • a machine learning model assesses the BG measurements and produces predictive hypoglycemic warnings at e.g. 24 hours, 15 hours, 12 hours, 3 hours, 2 hours, 60 minutes, 30 minutes or 15 minutes prediction horizons.
  • the hypoglycemic warnings may be communicated to the patient, e.g. via an application on a mobile phone, following rules dictated by the patient's risk profile.
  • the device is accessible within any browser (phone, tablet, laptop/desktop). All data used by the device to predict the risk of hypoglycemia may be accessible via a mobile device, server or cloud.
  • the device, system and method according to the invention builds on a machine learning approach to learn the individual patient's pattern and behavior in terms of historic BG levels and insulin injections.
  • the invention can automatically adjust a given
  • the standard titration algorithm (first algorithm) will not fall into a loop of recovering from the occasional hypoglycemic events, as the risk prediction module will foresee these events before they happen and correct the dose. This will reduce number of hypoglycemic events meaning less long term complications and result in better adherence for the patient.
  • the device and method according to the invention also is able to predict a risk of hyperglycemia (high BG levels) and adjust a first recommended insulin dose size to a second recommended dose size, in this case a second dose size being higher than the first dose size in order to reduce the BG level and prevent a hyperglycemic event.
  • Figure 1 illustrates an exemplary system topology of the modules included in the device for predicting risk of hypoglycemia and adjusting an insulin medicament dosage for a subject accordingly, in accordance with an embodiment of the present disclosure.
  • Figure 2 shows graphs of BG levels originally measured vs. BG levels predicted by a device for predicting a risk of hypoglycemia according to the invention.
  • Figures 3a-b illustrate data plots of BG levels and associated injected doses of basal insulin over time for a patient taking basal insulin doses computed according to the first algorithm and computed by a device/system/method according to the invention, respectively.
  • Figures 4a-b illustrate data plots of BG levels and associated injected doses of bolus insulin over time for a patient taking bolus insulin doses computed according to the first algorithm and computed by a device/system/method according to the invention, respectively.
  • the present disclosure relies upon the acquisition of data regarding a data set comprising a plurality of glucose measurements of a subject taken over a time course and, for each respective glucose measurement in the plurality of glucose measurements, a corresponding timestamp representing when in the time course the respective glucose measurement was made, said data preferably being acquired via a continuous glucose monitoring device (CGM) worn by the subject.
  • CGM continuous glucose monitoring device
  • the present disclosure also relies upon acquisition of data from one or more insulin pens used to apply a recommended dose of prescribed insulin to the subject. Each record comprises a timestamp also specifying an amount of injected insulin medicament and the type of insulin injected.
  • the device By recording the type of insulin the device receives information regarding the specific PK/PD (Pharmacokinetics/Pharmacodynamics) profiles of the given insulin, which is used to calculate the Insulin on Board (IoB) factor in the subject, a factor that may be used in predicting BG levels.
  • PK/PD Pulsacokinetics/Pharmacodynamics
  • IoB Insulin on Board
  • insulin pen an injection device suitable for applying discrete doses of insulin, and wherein the injection device is adapted for logging and communicating dose related data.
  • a first algorithm 100 is adapted to compute a recommendation for a first dose size to the patient, said first algorithm computing the dose size recommendation based upon a retrospective analysis of the historic BG and insulin injection data included in the first data set 110 (Patient's Metric).
  • the data in the first data set is fed into the device via a wireless receiver, where the BG measurements 120 is obtained wirelessly from a CGM affixed to the subject and the insulin injections data 130 is obtained wirelessly from one or more insulin pens adapted to communicate with the wireless receiver. For instance, in some embodiments the device receives this data wirelessly through radio-frequency signals.
  • such signals are in accordance with an 802.11 (WiFi), Bluetooth, or ZigBee standard.
  • the CGM and/or insulin injection pen includes an RFID tag and communicates to the device.
  • the first data set also includes physiological measurements of the subject (e.g., from wearable physiological measurement devices, etc).
  • the glucose sensor may e.g. be a FREESTYLE LIBRE CGM by ABBOTT ("LIBRE") used to make autonomous glucose measurements of a subject.
  • the LIBRE allows calibration- free glucose measurements with an on-skin coin-sized sensor, which can send up to eight hours of data to the device via near field communications or Bluetooth or any other suitable protocol.
  • Another example of a glucose sensor is DEXCOM G5.
  • the device is not proximate to the subject and/or does not have wireless capabilities or such wireless capabilities are not used for the purpose of acquiring glucose data, insulin medicament injection data, and/or physiological measurement data.
  • a communication network may be used to communicate data.
  • the device comprises, in addition the first algorithm, a risk prediction module 140 that receives input from a plurality of separate algorithm modules.
  • a risk prediction module 140 receives input from a plurality of separate algorithm modules.
  • One such module is the subject's pattern recognition algorithm module 150 that is adapted to analyze the correlations between the BG measurement data and insulin injection data of the first data set 110 in order to identify historic hyper- and hypoglycemic events over a predetermined period of time and the severity as to BG level of each of these events.
  • Another module is the short term risk prediction algorithm module 160 that is able to predict a risk of hypoglycemia occurring within a time period of up to 2 hours from start of computing the risk.
  • the risk prediction being computed based only on the BG measurements of the first data set. However, other data than the BG measurements may be used to compute the risk.
  • Yet another module is the long term risk prediction algorithm module 170 is able to predict a risk of hypoglycaemia occurring within a time period of up to 24 hours from start of computing the risk.
  • the risk prediction being computed by running a machine learning classification algorithm and based only on the BG measurements of said first data set. Other data than BG measurements may be used to compute the risk.
  • a multiple linear regression machine learning algorithm to predict a risk of hypoglycemia and calculates the second dose size being resulting in a non-hypoglycemic state. If the second dose size is lower than the dose first dose size calculated by the first algorithm, the risk prediction module communicates this new second dose size to the patient.
  • a further indirect measurements module 180 may be included containing sets of historic data of one or more of insulin sensitivity factor (ISF) measurements, Insulin on board (IoB) measurements, carbohydrates on board (CoB) measurements, and for each of the respective measurements a corresponding timestamp representing when in the time course the respective measurement, said second data set being used by the risk prediction module to predict a risk of hypoglycaemia.
  • This module 180 feeds data to the risk prediction module 140 to be used in predicting the risk and calculates a corrected insulin dose resulting in a non-hypoglycaemic state for the patient.
  • the first data set also comprises minimum and maximum targets for BG level and it may also comprise data about the insulin to carbohydrates ratio and potentially also body weight, as depicted in module 190 in figure 1.
  • a graph 200 showing BG levels originally measured in a patient is compared to a graph 210 showing BG levels predicted by the risk prediction module and with a 15 minutes risk prediction time horizon over a period of 24 hours.
  • the prediction is based on historic BG data only, and as can be seen, the risk prediction module is able to predict the BG levels with a very high accuracy, as there is a very little difference in BG values between the two graphs.
  • An often used value to represent the differences between predicted values and observed values model is the RMSE (Root Mean Square Error) value.
  • the RMSE value can be considered as a performance measure for the risk prediction model, and in the example shown the RMSE value is 2.0309 mg/dl for BG values ranging from 60-220 mg/dl, which is considered a very low RMSE representing a very good performance of by the risk prediction module.
  • Figure 3a illustrates a data plot of a patient's BG levels and associated injected dose sizes of basal and bolus insulin over time (horizontal axis indicates time).
  • the curve 300 shows a patient's BG profile over time for a patient taking basal (301, 303, 305, 307) and bolus (302, 304, 306, 308) insulin doses as computed and recommended according to first algorithm.
  • Figure 3b shows same kind of a data plot, but where curve 310 shows a patient's BG profile over time for a patient taking basal insulin doses (301, 303, 305, 307) and bolus insulin doses (302, 304, 306a, 308) as computed and recommended by a device/system/method according to the invention.
  • the lines 320 define the
  • normoglycemic state with BG ranges from 80-120 mg/dl, whereas the lines 330, 340 define the hypoglycemic state with BG ranges of 70 mg/dl and 60 mg/dl, respectively.
  • a patient may have changes in her/his normal routine on selected days during the week. There are many reasons for that, such as having a low carb meal or doing sports after the meal. Such changes are referred to by changes in bolus injections. As can be seen in Figure 3a, the BG level after injection of bolus dose 306 changes suddenly and gets below 70 mg/dl and the patient gets into a hypoglycemic state. The first algorithm is not able to predict such a periodic deviation, as it only acts retrospectively.
  • a titration algorithm is provided that is able to predict hypoglycemic event AND act proactively and
  • the patient will inject the second dose size 306a and not a first recommended dose size 306, whereby the BG level will not drop below 70 mg/dl (as depicted in the graph).
  • the patient is kept in the normoglycemic range and will not suffer from a hypo and the recovering from that. The patient will have fewer fluctuations in the BG level.
  • Figures 4a-b illustrate the same pattern as Figures 3a-b, here it is just basal dose adjustment instead of bolus dose adjustment.
  • the curve 400 shows a patient's BG profile over time for a patient taking basal (401, 403, 405, 407) and bolus (402, 404, 406, 408) doses as computed and recommended according to a standard insulin titration algorithm (referred to as the first algorithm).
  • Figure 4b shows same kind of a data plot, but where curve 410 shows a patient's BG profile over time for a patient taking basal insulin doses (401, 403a, 405, 407) and bolus insulin doses (402, 404, 406, 408) as computed and recommended by a device/method in according to the invention.
  • the lines 420 define the normoglycemic state with BG ranges from 80-120 mg/dl, whereas the lines 430, 440 define the hypoglycemic state with BG ranges of 70 mg/dl and 60 mg/dl, respectively.
  • the patient may have changes in her/his normal routine on selected days during the week.
  • the patient's' fasting BG level goes below 60 mg/dl after basal injection 403 and bolus injection 404 meaning the patient just after the fasting period is in a hypoglycemic state. It may be due to the fact that the patient is having heavy physical activity or a low carb dinner meal that specific evening or a fasting day in the week.
  • the first algorithm is not able to predict such periodic deviation proactively, as it only acts retrospectively.
  • the hypoglycemic state after basal injection dose 403 in Figure 4a has been predicted by the risk prediction module according to the invention, i.e. the low BG level is predicted and avoided as the risk prediction module proactively and automatically calculates a new second lower basal dose 403a and inputs this dose to the first algorithm and thereby overrules the first recommended dose 403.
  • the patient will inject the second dose size 403a and not a first recommended dose size 403, whereby the BG level will not drop below 70 mg/dl (as depicted in the graph).
  • the patient is kept in the normoglycemic range during fasting period.
  • the present invention can be implemented as a computer program product that comprises a computer program mechanism embedded in a non-transitory computer readable storage medium.
  • the computer program product could contain the modules shown in any combination of Figures 1. These modules can be stored on a CD-ROM, DVD, magnetic disk storage product, or any other non-transitory computer readable data or program storage product.

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Public Health (AREA)
  • Medical Informatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Pathology (AREA)
  • Medicinal Chemistry (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Chemical & Material Sciences (AREA)
  • Investigating Or Analysing Biological Materials (AREA)

Abstract

La présente invention concerne un dispositif, un système et un procédé permettant de prédire le risque d'hypoglycémie et de régler la dose de médicament à base d'insuline destinée à un sujet nécessitant des injections par stylo plusieurs fois par jour. Un premier algorithme sert à calculer une première dose d'insuline en fonction d'une pluralité de mesures de glycémie (BG) horodatées et de doses d'insuline injectées. Un module de prédiction de risques est configuré, sur la base d'algorithmes d'apprentissage automatique, pour prédire le risque d'hypoglycémie sur la base des mesures de BG horodatées et pour calculer une seconde dose d'insuline conduisant à un état de non-hypoglycémie. Dans le cas où la seconde dose est inférieure à la première dose, la première dose est automatiquement diminuée par réglage sur la seconde dose et communiquée comme la dose recommandée au sujet.
PCT/EP2018/065845 2017-06-15 2018-06-14 Algorithme de dosage d'insuline sur la base d'un profil de patient WO2018229209A1 (fr)

Priority Applications (4)

Application Number Priority Date Filing Date Title
JP2019568745A JP7181900B2 (ja) 2017-06-15 2018-06-14 患者プロファイルに基づくインスリン滴定アルゴリズム
US16/622,384 US20200227170A1 (en) 2017-06-15 2018-06-14 Insulin titration algorithm based on patient profile
EP18729473.1A EP3639168B1 (fr) 2017-06-15 2018-06-14 Algorithme de titrage de l'insuline basé sur le profil du patient
CN201880039363.2A CN110753967A (zh) 2017-06-15 2018-06-14 基于患者概况的胰岛素滴定算法

Applications Claiming Priority (6)

Application Number Priority Date Filing Date Title
US201762520139P 2017-06-15 2017-06-15
US62/520,139 2017-06-15
US201762522811P 2017-06-21 2017-06-21
US62/522,811 2017-06-21
EP17178877.1 2017-06-29
EP17178877 2017-06-29

Publications (1)

Publication Number Publication Date
WO2018229209A1 true WO2018229209A1 (fr) 2018-12-20

Family

ID=59258137

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/EP2018/065845 WO2018229209A1 (fr) 2017-06-15 2018-06-14 Algorithme de dosage d'insuline sur la base d'un profil de patient

Country Status (1)

Country Link
WO (1) WO2018229209A1 (fr)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021165747A1 (fr) 2020-02-17 2021-08-26 GlucoGear Tecnologia LTDA Dispositifs, systèmes et procédés de prédiction de taux de glucose dans le sang sur la base d'un modèle de régulation de glucose dans le sang personnalisé
CN116721733A (zh) * 2023-08-10 2023-09-08 武汉联影智融医疗科技有限公司 血糖值调整方法、装置和存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6923763B1 (en) 1999-08-23 2005-08-02 University Of Virginia Patent Foundation Method and apparatus for predicting the risk of hypoglycemia
WO2011157402A1 (fr) 2010-06-18 2011-12-22 Roche Diagnostics Gmbh Systèmes d'optimisation de l'insuline et procédés de test avec critère de sortie ajusté, prenant en considération le bruit du système associé aux biomarqueurs
US8562587B2 (en) 2009-02-25 2013-10-22 University Of Virginia Patent Foundation CGM-based prevention of hypoglycemia via hypoglycemia risk assessment and smooth reduction of insulin delivery
US20140350369A1 (en) 2013-03-15 2014-11-27 Abbott Diabetes Care Inc. System and method to manage diabetes based on glucose median, glucose variability, and hypoglycemic risk
US20150190098A1 (en) 2011-08-26 2015-07-09 Stephen D. Patek Method, System and Computer Readable Medium for Adaptive and Advisory Control of Diabetes
US20170068790A1 (en) 2014-02-28 2017-03-09 Eco-Fusion Systems for predicting hypoglycemia and methods of use thereof

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6923763B1 (en) 1999-08-23 2005-08-02 University Of Virginia Patent Foundation Method and apparatus for predicting the risk of hypoglycemia
US8562587B2 (en) 2009-02-25 2013-10-22 University Of Virginia Patent Foundation CGM-based prevention of hypoglycemia via hypoglycemia risk assessment and smooth reduction of insulin delivery
WO2011157402A1 (fr) 2010-06-18 2011-12-22 Roche Diagnostics Gmbh Systèmes d'optimisation de l'insuline et procédés de test avec critère de sortie ajusté, prenant en considération le bruit du système associé aux biomarqueurs
US20150190098A1 (en) 2011-08-26 2015-07-09 Stephen D. Patek Method, System and Computer Readable Medium for Adaptive and Advisory Control of Diabetes
US20140350369A1 (en) 2013-03-15 2014-11-27 Abbott Diabetes Care Inc. System and method to manage diabetes based on glucose median, glucose variability, and hypoglycemic risk
US20170068790A1 (en) 2014-02-28 2017-03-09 Eco-Fusion Systems for predicting hypoglycemia and methods of use thereof

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021165747A1 (fr) 2020-02-17 2021-08-26 GlucoGear Tecnologia LTDA Dispositifs, systèmes et procédés de prédiction de taux de glucose dans le sang sur la base d'un modèle de régulation de glucose dans le sang personnalisé
CN116721733A (zh) * 2023-08-10 2023-09-08 武汉联影智融医疗科技有限公司 血糖值调整方法、装置和存储介质
CN116721733B (zh) * 2023-08-10 2023-11-07 武汉联影智融医疗科技有限公司 血糖值调整方法、装置和存储介质

Similar Documents

Publication Publication Date Title
EP3639168B1 (fr) Algorithme de titrage de l'insuline basé sur le profil du patient
US20210050085A1 (en) Systems, devices, and methods relating to medication dose guidance
CN107809943B (zh) 用于分析从患有糖尿病的人测得的葡萄糖数据的方法和系统
CN114796704B (zh) 用于胰岛素的递送的设备和方法
US20210193285A1 (en) A system and method for use in disease treatment management
CN104837517B (zh) 用于调谐人工胰腺的闭环控制器的方法和系统
US9990581B2 (en) Insulin dosage assessment and recommendation system
KR20190132684A (ko) 인공 췌장 시스템에서의 잔존 인슐린 고려
US20140128705A1 (en) Computer-based diabetes management
US11195607B2 (en) Starter kit for basal rate titration
US11282598B2 (en) Starter kit for basal insulin titration
CN109661706B (zh) 用于确定胰岛素敏感性的系统和方法
US20230005587A1 (en) Determining whether adjustments of insulin therapy recommendations are being taken into account
US20230005586A1 (en) Determining total daily basal dose mismatch
WO2018229209A1 (fr) Algorithme de dosage d'insuline sur la base d'un profil de patient
US20210151141A1 (en) Joint state estimation prediction that evaluates differences in predicted vs. corresponding received data
Rodríguez-Rodríguez et al. Forecasting glycaemia for type 1 diabetes mellitus patients by means of IoMT devices

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18729473

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2019568745

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2018729473

Country of ref document: EP

Effective date: 20200115