Computer implemented method of training a machine learning model configured for predicting a hypoglycemic event for a subject

Abstract

A computer implemented method of training a machine learning model configured for predicting a hypoglycemic event for a subject is proposed, comprising: i. (126) receiving at least two time series of glucose sensor data, wherein each time series of glucose sensor data comprises a plurality of glucose measurements measured by a glucose monitoring device (112) configured for detecting glucose in a bodily fluid of the subject, wherein a first time series of the two time series of glucose sensor data is measured at least during and / or after being exposed to a first predefined program of glycemic stimuli, and wherein a second time series of the two time series of glucose sensor data is measured at least during and / or after being exposed to a second predefined program of glycemic stimuli, wherein the second predefined program of glycemic stimuli is different from the first predefined program of glycemic stimuli; ii. (128) forming a first training dataset using the first time series and forming a second training dataset using the second time series; iii. (130) training the machine learning model with the first training dataset and the second training dataset.

Description

[0001] P39298

[0002] - 1 -

[0003] November 28, 2025

[0004] Computer implemented method of training a machine learning model configured for predicting a hypoglycemic event for a subject

[0005] Technical Field

[0006] The invention relates to a computer-implemented method of training a machine learning model configured for predicting a hypoglycemic event for a subject, a computer implemented method for managing a predicted hypoglycemic event, a continuous glucose monitoring system, a computer program and a computer-readable storage medium. The invention may both be applied in the field of home care as well as in the field of professional care, such as in hospitals. Other applications are feasible.

[0007] Background art

[0008] Several approaches for prediction of glycemic events such as hypoglycemic events from Continuous Glucose Monitoring (CGM) data based on machine learning are known. Such machine learning method usually are based on machine learning solutions which were trained with population user data or which adapt based on the physiological data and user input provided by the specific user.

[0009] Hitherto known machine learning methods in this area suffer from the fact that the systems either do not adapt to the specific user or if they do, they have to deal with imperfect user data due to the fact that user enter incomplete of wrong data such as meal data, administered insulin data, physical activity data, etc. which in turn adversely affects the machine learning model’s ability to learn from the recorded CGM values that lack adequate context information so as to improve is glucose prediction performance.P39298

[0010] - 2 -

[0011] US 2021 / 0256872 Al describes devices, systems, and methods for optimizing blood glucose level regulation by predicting blood glucose levels based on personalized blood glucose regulation models. In some exemplary embodiments, a computer-implemented method of blood glucose level regulation includes collecting a first plurality of data sets associated with an individual from a database, generating a personalized blood glucose regulation model for the individual, receiving a second plurality of data sets associated with the individual, and generating predicted blood glucose levels for the individual using the personalized blood glucose regulation model and the second plurality of data sets.

[0012] US 2021 / 233656 Al describes a method predicting a predetermined health condition of a user based on a personalized user glucose prediction, vital sign, and motion data detected by a wearable device. Based on such a prediction a plan is generated, e.g. to control user glucose by controlling diet and physical activity. The method aims, in particular, to provide reduced hypoglycemic events and time in hypoglycemia.

[0013] US 2024 / 324965 Al discloses systems and methods for biomonitoring and personalized healthcare. In some embodiments, a computer-implemented method for forecasting a blood glucose state of a patient is provided. The method comprises: receiving blood glucose data of the patient; generating at least one initial prediction of the blood glucose state by inputting the blood glucose data into a first set of machine learning models; determining a plurality of features at least partly from the at least one initial prediction; and generating a final prediction of the blood glucose state by inputting the plurality of features into a second set of machine learning models.

[0014] Problem to be solved

[0015] It is therefore desirable to a computer-implemented method of training a machine learning model configured for predicting a hypoglycemic event for a subject, a computer implemented method for managing a predicted hypoglycemic event, a continuous glucose monitoring system, a computer program and a computer-readable storage medium, which solve at least one of the problems mentioned above. Specifically, devices and methods shall be proposed which improve training of the machine learning model, which allows an improved performance in predicting glycemic events.

[0016] SummaryP39298

[0017] - 3 -

[0018] This problem is addressed by a computer-implemented method of training a machine learning model configured for predicting a hypoglycemic event for a subject, a computer implemented method for managing a predicted hypoglycemic event, a continuous glucose monitoring system, a computer program and a computer-readable storage medium with the features of the independent claims. Advantageous embodiments which might be realized in an isolated fashion or in any arbitrary combinations are listed in the dependent claims as well as throughout the specification.

[0019] As used in the following, the terms “have”, “comprise” or “include” or any arbitrary grammatical variations thereof are used in a non-exclusive way. Thus, these terms may both refer to a situation in which, besides the feature introduced by these terms, no further features are present in the entity described in this context and to a situation in which one or more further features are present. As an example, the expressions “A has B”, “A comprises B” and “A includes B” may both refer to a situation in which, besides B, no other element is present in A (i.e. a situation in which A solely and exclusively consists of B) and to a situation in which, besides B, one or more further elements are present in entity A, such as element C, elements C and D or even further elements.

[0020] Further, it shall be noted that the terms “at least one”, “one or more” or similar expressions indicating that a feature or element may be present once or more than once typically will be used only once when introducing the respective feature or element. In the following, in most cases, when referring to the respective feature or element, the expressions “at least one” or “one or more” will not be repeated, non-withstanding the fact that the respective feature or element may be present once or more than once.

[0021] Further, as used in the following, the terms "preferably", "more preferably", "particularly", "more particularly", "specifically", "more specifically" or similar terms are used in conjunction with optional features, without restricting alternative possibilities. Thus, features introduced by these terms are optional features and are not intended to restrict the scope of the claims in any way. The invention may, as the skilled person will recognize, be performed by using alternative features. Similarly, features introduced by "in an embodiment of the invention" or similar expressions are intended to be optional features, without any restriction regarding alternative embodiments of the invention, without any restrictions regarding the scope of the invention and without any restriction regarding the possibility of combining the features introduced in such way with other optional or non-optional features of the invention.P39298

[0022] - 4 -

[0023] In a first aspect, a computer implemented method of training a machine learning model configured for predicting a hypoglycemic event for a subject is disclosed.

[0024] The method comprises the following method steps, which, as an example, may be performed in the given order. However, a different order is also feasible. Further, it is possible to perform two or more of the method steps simultaneously or in a fashion overlapping in time. Further, it is also possible to perform one, more than one or even all of the method steps repeatedly.

[0025] The method comprises the following steps:

[0026] i. receiving at least two time series of glucose sensor data, wherein each time series of glucose sensor data comprises a plurality of glucose measurements measured by a glucose monitoring device configured for detecting glucose in a bodily fluid of the subject,

[0027] wherein a first time series of the two time series of glucose sensor data is measured at least during and / or after being exposed to a first predefined program of glycemic stimuli, and

[0028] wherein a second time series of the two time series of glucose sensor data is measured at least during and / or after being exposed to a second predefined program of glycemic stimuli, wherein the second predefined program of glycemic stimuli is different from the first predefined program of glycemic stimuli;

[0029] ii. forming a first training dataset using the first time series and forming a second training dataset using the second time series;

[0030] iii. training the machine learning model with the first training dataset and the second training dataset.

[0031] The present invention proposes using predefined glycemic stimulus exposure programs of different categories such as carbohydrate intake, insulin administration, and physical activity to generate user specific training data. This can allow training the machine learning model based on personalized training data, preferably comprising data for all three categories, i.e. carbohydrate intake, insulin administration, and physical activity. This can significantly enhance the quality of the personalized training data and prediction accuracy for predicting a hypoglycemic event.

[0032] For example, a specific user is subjected to a complex standardized and predefined glycemic stimulus exposure program, possibly under monitoring / control of a health care professional. The resulting glucose measurements may be recorded during execution of the program(s).P39298

[0033] - 5 -

[0034] The predefined glycemic stimulus exposure program(s) may comprise defined meal con-sumption / deprivation exercises, physical activity exercises, insulin administration / depriva-tion exercises, and the like. This can allow generating glucose measurement and, optionally other physiological data, that reflects the subject’s behavior in response to the predefined glycemic stimulus exposure. The machine learning model, such as population user data trained models, which is used for predicting hypoglycemic events can then be better personalized for the specific user based on the recorded specific user data which was collected during the glycemic stimulus exposure program(s).

[0035] The term "computer implemented" as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a method which is performed by using computer programming, and / or by using at least one computer and / or at least one computer network. Thus, as an example, one or more or even all of the method steps may be performed by appropriate software, e.g. by using computer-readable instructions which, when executed on a computer or a computer network, cause the computer or computer network to perform the method steps. The term “software” as used herein may, specifically, refer to a computer program. The computer program may have a plurality of functions, procedures, methods and subprograms, which may be distributed over several specific hardware instances. The computer and / or computer network may comprise at least one processor, which is configured for performing at least one, more than one or all of the method steps of the method according to the present disclosure. The computer and / or computer network may comprise at least one memory configured for storing instruction, such as instructions related to the computer implemented method. Specifically, each of the method steps is performed by the computer and / or computer network.

[0036] The method may be performed completely automatically. The term "automatically" as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a process which is performed completely by means of at least one computer and / or computer network and / or machine, in particular without manual action and / or interaction with a user.

[0037] The term “hypoglycemic event” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to oc-P39298

[0038] - 6 -

[0039] currence of a hypoglycemic condition of the subject. The hypoglycemic event may be associated with a glucose concentration being below a predetermined threshold value for at least a predetermined time interval. For example, the predetermined threshold value may refer to a glucose level below 70 mg / dl. However, other thresholds are possible. A hypoglycemic event can occur at any time of the day, e.g. during night such as during sleep time of the subject. A hypoglycemic event occurring at night may be denoted as nocturnal hypoglycemia.

[0040] The term “subject” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a vertebrate animal, in an embodiment a mammal and, in a further embodiment, to a human. In an embodiment, the subject is known or suspected to suffer from diabetes, in an embodiment diabetes type II or type I and / or shows signs and / or symptoms of diabetes, hyperglycemia, and / or hypoglycemia, which are known from medical textbooks. Suspicion to suffer from diabetes may in particular stem from preceding diagnostic measures, such as anamnesis, physical examination, clinical chemistry diagnostics, in particular blood glucose measurement, and the like.

[0041] The term “predicting a hypoglycemic event” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to forecasting if a hypoglycemic event is likely to occur for the subject. The prediction may be a qualitative prediction, i.e. expected occurrence of a hypoglycemic event or expected non-occurrence of a hypoglycemic event. Additionally or alternatively, the prediction may comprise a quantitative prediction, e.g. indication likelihood of occurrence of a hypoglycemic event, e.g. in percent.

[0042] The term “machine learning” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a method of using artificial intelligence (Al) for automatically model building of machine learning models, in particular of prediction models. For example, the machine learning may comprise deep learning. The term “deep learning” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a class of machine learning algorithms using multiple layers, in particular usingP39298

[0043] - 7 -

[0044] deep learning architectures such as one or more of deep neural networks, deep belief networks, graph neural networks, recurrent neural networks and convolutional neural networks.

[0045] The term “machine learning model” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a trainable model, wherein the training comprises and / or is based on and / or uses machine learning. The trainable model may be or may comprise at least one mathematical model configured for transforming one or more input values into one or more output values by using one or more parameters which may be adjusted in order to enable the model to be trained. The trainable model specifically may be a trainable mathematical model which is trainable on at least one training dataset using one or more of machine learning, deep learning, neural networks, or other form of artificial intelligence. The term “trainable model” specifically may refer, without limitation, to the fact that the trainable model can be further trained, optimized or updated based on training data, e.g. of a training dataset. The trainable model may be trained by using machine learning.

[0046] The term “training” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a process of determining parameters of at least one machine learning model, specifically of the algorithm of the machine learning model, specifically on at least one training dataset or set of training data. The training specifically may comprise at least one optimization or tuning process, wherein a best parameter combination, e.g. according to at least one optimization procedure, is determined. The term “optimization”, as used herein, is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to the process of selecting of a best parameter set with regard to the optimization target from a parameter space of possible parameters. The term “optimization target”, as used herein, is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to at least one criterion under which the optimization is performed. The optimization target may comprise at least one optimization goal and accuracy and / or precision.

[0047] The machine learning model may be a pretrained model. The term “pretrained model” as used herein is a broad term and is to be given its ordinary and customary meaning to a personP39298

[0048] - 8 -

[0049] of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to the fact that the machine learning model was subject to at least one training process. Thus, the training of the machine learning model may refer to fine-tuning of the machine learning model. The term “fine-tuning” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to training the parameters of the pretrained model on additional data which was not used for pre-training the model. For example, the machine learning model is pretrained using historical time series of glucose measurements of one or more of: labelled data of a user population, labelled data of the subject, labelled data of at least one study, further knowledge. The pretraining of the machine learning model may comprise using at least one classifier selected from the group consisting of: Linear Classifier, Logistic Regression, Naive Bayes Classifier, Perceptron, Support Vector Machine; Quadratic Classifier, K-Means Clustering, Boosting, Decision Tree, Random Forest; Neural network, Bayesian Network and the like. For example, the pretrained model may be generated as described in Jensen et al. “Prediction of Nocturnal Hypoglycemia From Continuous Glucose Monitoring Data in People With Type 1 Diabetes: A Proof-of-Concept Study”, Journal of Diabetes Science and Technology 2020, Vol. 14(2) 250- 256.

[0050] As outlined above, step i) comprises receiving at least two time series of glucose sensor data. Each time series of glucose sensor data comprises a plurality of glucose measurements measured by a glucose monitoring device configured for detecting glucose in a bodily fluid of the subject.

[0051] The term "detecting" as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a quantitative and / or qualitative determination by using at least one sensor.

[0052] The term “receive” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to the process of obtaining data from a data source. The data source may, however, vary, in accordance with the specific application. For example, the receiving may comprise at least one of the following: downloading from at least one data source, such as from at least one data storage device or from a web- or cloud-based data storage device; obtaining via at least one computer network, such as the Internet; obtaining via at least one wire-based and / or wireless interface.P39298

[0053] - 9 -

[0054] The receiving may fully or partially take place automatically, such as by automatic download, and / or may fully or partially take manually. Semi-automatic receiving processes may be also possible. For example, the time series of glucose sensor data may be generated by the glucose monitoring device and may be transmitted, e.g. to at least one processor for training the machine learning model, e.g. directly or by using at least one data storage device or a web- or cloud-based data storage device.

[0055] For example, the receiving may comprise performing at least one measurement, in particular a plurality of measurements by using the glucose monitoring device. The term “glucose monitoring device” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term preferably may refer, without limitation, to a device comprising at least one sensor configured for qualitatively or quantitatively detecting the presence and / or the concentration of glucose in bodily fluid. As will be understood by the skilled person, a "presence" may be a presence of glucose in an amount above a detection limit. The glucose monitoring device may comprise an assembly of two or more components capable of interacting with each other, such as in order to perform a detection of glucose in the body fluid and / or of contributing to the detection of glucose in the body fluid. The glucose monitoring device generally may also be or may comprise at least one of a sensor assembly, a sensor system, a sensor kit or a sensor device.

[0056] The term "bodily fluid" as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term may specifically refer, without limitation, to a fluid, in particular a liquid, which is typically present in a body or a body tissue of the subject and / or may be produced by the body of the subject. For example, the bodily fluid may be selected from the group consisting of blood and interstitial fluid. However, additionally or alternatively, one or more other types of bodily fluids may be used, such as saliva, tear fluid, urine or other body fluids. The glucose monitoring device may be configured for being at least partially inserted into a body tissue of the user. The term “body tissue” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term preferably may refer, without limitation, to a cellular organizational level intermediate between cells and a complete origin. The body tissue may specifically be an ensemble of similar cells from the same origin that together carry out a specific function. Thereby, organs may thenP39298

[0057] - 10 -

[0058] be formed by functional grouping together of multiple tissues. As an example for body tissue, interstitial tissue, i.e. connective tissue between cellular elements if a structure, may be named.

[0059] The glucose monitoring device may be configured for continuous glucose monitoring (CGM). The term “continuous glucose monitoring” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to the fact that the glucose monitoring device or at least a part of the glucose monitoring device e may remain in the body tissue of the user for a predetermined period of time, such as for several hours, specifically for one or more days, more specifically for up to one week, even more specifically for up to two weeks or even more. Example of such known glucose monitoring device comprise the Dexcom G7 CGM, the Abbott Freestyly Libre 3, and the Roche Accu-Chek SmartGuide CGM device.

[0060] The term “data” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to at least one discrete or continuous value that comprises information. Typically, data may comprise a plurality of said discrete or continuous values. The term “glucose sensor data” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to data obtained by using the glucose monitoring device, e.g. a raw sensor signal, a processed sensor signal and / or information derived from the sensor signal. For example, the glucose sensor data may comprise a glucose concentration value, e.g. a molar concentration value or a mass concentration value. The term “time series of glucose sensor data” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a sequence of glucose measurements, which are ordered with respect to measurement time, e.g. starting with the oldest glucose measurement. Each of the time series may comprise a time-ordered sequence of the glucose measurements. Each of the glucose measurements may comprise a time stamp. The term “time stamp” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to information about the time point of generating the glucose measurement. The time stamp may comprise information about a date and a time of day e.g. with a resolution ofP39298

[0061] - 11 -

[0062] seconds. The time stamp may be assigned to the corresponding glucose measurement by a processor, e.g. of the glucose monitoring device.

[0063] The first time series of the two time series of glucose sensor data is measured at least during and / or after being exposed to a first predefined program of glycemic stimuli. The second time series of the two time series of glucose sensor data is measured at least during and / or after being exposed to a second predefined program of glycemic stimuli. The second predefined program of glycemic stimuli is different from the first predefined program of glycemic stimuli.

[0064] The term “glycemic stimulus” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to at least one action that influences and / or affects the glucose level of the subject. The term “predefined program of glycemic stimuli” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a program comprising at least one predefined step of exposing the subject to at least one glycemic stimulus. For example, the predefined program of glycemic stimuli may be standardized, e.g. in the sense that for each subject or a subject of a specific category, e.g. age, disease state and the like, an identical predefined program of glycemic stimuli is used. Alternatively, the predefined program of glycemic stimuli may be adapted for each subject. For example, the predefined program of glycemic stimuli may be defined by a health care professional. The exposure to the glycemic stimuli may be performed under monitoring and / or control of a health care professional.

[0065] The predefined glycemic stimuli exposure program may comprise at least one arbitrary stimulus expected to result in a change in glucose level and / or expected to have an influence on the subject’s glucose level. For example, each of the predefined programs comprises at least one quantitatively and qualitatively predefined glycemic stimulus which the subject is exposed to for a predefined time period. For example, each of the predefined programs comprises a series of quantitatively and qualitatively predefined glycemic stimuli which the subject is exposed to for a predefined time period. For example, the glycemic stimuli comprise at least one stimulus selected from the group consisting of: at least one defined meal consumption; at least one food deprivation exercise, at least one physical activity exercise; at least one insulin administration; at least one drug deprivation exercise.P39298

[0066] - 12 -

[0067] The first predefined program may comprise at least one glycemic stimulus of a first category and the second predefined program may comprise at least one glycemic stimulus of a second category. The first and the second category may differ. For example, the categories are selected from physical activity, food consumption, food deprivation, drug consumption, drug deprivation, insulin administration or insulin deprivation. For example, the first predefined program of glycemic stimuli comprises consuming a defined type and amount of carbs at defined day times. For example, the second predefined program of glycemic stimuli comprises performing a defined amount of physical activity over a defined time interval and / or at defined day times.

[0068] For example, the method may comprise receiving at least one glucose measurement, in particular a first time series of glucose sensor data, from the glucose monitoring device before the subject being exposed to a first and / or second predefined program of glycemic stimuli. These data may be used for comparison with glucose sensor data received during and / or after the subject being exposed to a first or second predefined program of glycemic stimuli.

[0069] The method may comprise detecting, in addition to the time series of glucose sensor data measured during and / or after the exposure to the predefined glycemic stimuli, at least one physiological value of the subject in response to the respective stimulus, e.g. physiological data that reflect the subject’s behavior in response to the predefined glycemic stimulus exposure. Said physiological value may be recorded during and / or after the execution of the first and / or second predefined program of glycemic stimuli. The physiological value of the subject may be one or more of user physical activity data from a sensor such as from a motion sensor or a heart rate sensor, glucose sensor data from the glucose monitoring device, user meal consumption data, user’s past drug administration data, drug-on-board or drug metabolite-on-board data, or a user’s disease state data. The user may be the subject. The term “motion sensor” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a device configured for detecting motion, in particular of at least one part of the body of the subject. For example, the motion sensor may be configured for detecting the motion sensor's intrinsic motion or the motion of another device or obj ect. The motion sensor may be or may comprise at least one of an angular rate sensor; a gyroscope; an accelerometer; a microelectromechanical systems (MEMS) accelerometer; or an inertial measurement unit. The term “heart rate sensor” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customizedP39298

[0070] - 13 -

[0071] meaning. The term specifically may refer, without limitation, to a device configured for detecting a heart rate. For example, the heart rate sensor may be or may comprise one or more of at least one photoplethysmogram (PPG) device or at least one electrocardiogram (ECG) device. For example, the user meal consumption data may comprise data relating to food intake such as one or more of meal start, meal end, amount of food, type of food and the like. The user meal consumption data may be input manually by the user e.g. by using at least one user interface, and / or may be determined at least partially automatically, e.g. by using at least one software application on a subject’s mobile device. For example, the user’s past drug administration data may comprise one or more of time of drug administration, duration of drug administration, amount of drug administration. The user’s past drug administration data may be input manually by the user e.g. by using at least one user interface, and / or may be determined at least partially automatically, e.g. by retrieving drug administration data from an infusion device. For example, the drug-on-board or drug metabolite-on-board data may comprise information on drug level, e.g. before and after administration. For example, the user’s disease state data may comprise arbitrary health data, in particular which can be used for and / or which have an influence on diabetes management e.g. change of disease state due to the at least one stimulus. The physiological value of the subject may be used as additional input for training of the machine learning model.

[0072] As outlined above, step ii. comprises forming a first training dataset using the first time series and forming a second training dataset using the second time series.

[0073] The term “training dataset” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a dataset comprising the training data on which the machine learning model is trained. The term “forming” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to generating the training dataset. The first time series of glucose sensor data may be used as data for the first training dataset. The second time series of glucose sensor data may be used as data for the second training dataset. The first training dataset may comprise information about the first predefined program of glycemic stimuli and / or the physiological value of the subject in response to the respective stimulus. The second training dataset may comprise information about the second predefined program of glycemic stimuli and / or the physiological value of the subject in response to the respective stimulus. The first and / or second training dataset may comprise additional information on the subject such as one or more of age, health data, activity data,P39298

[0074] - 14 -

[0075] and the like. The forming of the training dataset may comprise bringing the received data in a data format which can be used for training the machine learning model. The forming of the training dataset may comprise at least one preprocessing step, e.g. comprising smoothing, filtering and the like.

[0076] As outlined above, step iii. comprises training the machine learning model with the first training dataset and the second training dataset. The training with the first training dataset and the second training dataset may be performed at the same time, e.g. in parallel, or subsequently. For example, the machine learning model may be firstly trained on the first training dataset and, subsequently, may be trained on the second training dataset. In this case, the method may comprise receiving, in step i., the first time series only, forming the first training dataset in step ii. and training the machine learning model on the first training dataset in step iii.. Subsequently, steps i. to iii. may be performed for the second time series. Other embodiments and orders of steps may be possible. The terms “first” and “second” may refer to names only and give no information on an order or if additional time series are present.

[0077] The training of the machine learning model may comprise fine-tuning the pretrained machine model using the first training dataset and the second training dataset. The additional training of the pretrained machine learning model, such as a population user data trained model, can allow personalizing the pretrained machine learning model for the specific subject. This can be possible by training using the recorded specific user data which was collected during the predefined programs of glycemic stimuli.

[0078] The machine learning model may be trained using at least one classifier selected from the group consisting of: Linear Classifier, Logistic Regression, Naive Bayes Classifier, Perceptron, Support Vector Machine; Quadratic Classifier, K-Means Clustering, Boosting, Decision Tree, Random Forest; Neural network, Bayesian Network and the like.

[0079] For example, the training, in particular the fine-tuning, comprises supervised learning. The term “supervised” learning as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to learning based on labeled training data. The term “labeled” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a property of training data having, besides the actual training data, addi-P39298

[0080] - 15 -

[0081] tional information indicating a classification of the data. The labeled training data may comprise information about the glycemic stimuli the subject was exposed to. The labeled training data may comprise information about one or more of: information about a glucose level development within a predefined time range during and / or after exposing to the glycemic stimuli, information about occurrence of a hypoglycemic event within a predefined time range during and / or after exposing to the glycemic stimuli. For example, the first training dataset may comprise the first time series of glucose sensor data and a label which indicates whether or not a hypoglycemic event occurred. For example, the second training dataset may comprise the second time series of glucose sensor data and a label, which indicates whether or not a hypoglycemic event occurred. For example, the first training dataset may comprise information about the exposed first predefined program of glycemic stimuli and the second training dataset may comprise information about the exposed second predefined program of glycemic stimuli. Additionally or alternatively, the first training dataset and the second training dataset may comprise at least one determined physiological value.

[0082] The training method specifically may comprise splitting the training dataset into a training dataset and a test dataset. Thus, more specifically, the splitting into a training dataset and a test dataset may comprise x%-y% splitting of the training dataset, with x% being in the range 60 to 90, specifically in the range 65 to 75, and more specifically x%=70, and with y%=100-x. Therein, x% denotes the training dataset and y% denotes the test dataset.

[0083] Steps ii. and iii. may be carried out during and / or after data recording in step i..

[0084] The method may be used for predicting a nocturnal hypoglycemic event, in particular for predicting a hypoglycemic event occurring during a time the subject is sleeping. At least step i. may be performed during a predetermined time interval before, during and / or after a sleep time interval. The sleep time interval may be defined by user entry, e.g. via at least one user interface, and / or determined based on historical data collected with a sleep sensor, such as a sleep sensor configured for polysomnography or actigraphy. The training in step iii. may be carried out at defined time points, e.g. before sleep. The physiological value may be collected before and / or during and / or after exposing the subject to the glycemic stimuli including the time period during sleep. The method may further comprise at least one step of determining a medical risk associated with the predicted hypoglycemic event, e.g. during the sleep time interval. The method may further comprise outputting a notification to a display device to inform the subject and / or a further person, e.g. a health care professional, of the predicted hypoglycemic event if the medical risk during the sleep period is above a reference threshold.P39298

[0085] - 16 -

[0086] The method may comprise receiving a first time series of glucose sensor data from the glucose monitoring device before and / or during and / or after being exposed to the first predefined program of glycemic stimuli. For example, the first predefined program of glycemic stimuli may comprise consuming a defined type and amount of carbs at defined time points. The glucose measurement values may be recorded before and / or until and / or after the consumption and may constitute the first time series of glucose sensor data. The method may comprise training the machine learning model to associate the first time series of glucose sensor data with the first predefined program of glycemic stimuli.

[0087] The method may comprise receiving a second time series of glucose sensor data from the glucose monitoring device before and / or during and / or after being exposed to the second predefined program of glycemic stimuli. For example, the second predefined program of glycemic stimuli may comprise performing a defined amount of physical activity over a defined time interval. The glucose measurement values may be recorded before and / or until and / or after the activity constitute the second time series of glucose sensor data. The method may comprise training the machine learning model to associate the second time series of glucose sensor data with the second predefined program of glycemic stimuli.

[0088] The method may further comprises

[0089] iv. receiving at least one further time series of glucose sensor data, wherein each of the further time series of glucose sensor data comprises a plurality of glucose measurements measured by the at least one glucose monitoring device configured for detecting glucose in a bodily fluid of the subject, wherein the further time series is measured at least during and / or after being exposed to a further predefined program of glycemic stimuli, wherein the further predefined program of glycemic stimuli is different from the first and second predefined program of glycemic stimuli;

[0090] v. forming further training dataset using the further time series;

[0091] vi. training the machine learning model using the further training dataset.

[0092] The method may comprise receiving a third time series of glucose sensor data from the glucose monitoring device before and / or during and / or after being exposed to the third predefined program of glycemic stimuli. For example, the third predefined program of glycemic stimuli may comprise administering a defined amount of insulin at defined time points. The glucose measurement values may be recorded before and / or until and / or after the activity constitute the third time series of glucose sensor data. The method may comprise training the machine learning model to associate the third time series of glucose sensor data with the third predefined program of glycemic stimuli.P39298

[0093] - 17 -

[0094] In a further aspect, a computer implemented method for managing a predicted hypoglycemic event is disclosed. The method comprises the following method steps, which, as an example, may be performed in the given order. However, a different order is also feasible. Further, it is possible to perform two or more of the method steps simultaneously or in a fashion overlapping in time. Further, it is also possible to perform one, more than one or even all of the method steps repeatedly.

[0095] The method comprises:

[0096] a. receiving, by a processor, glucose sensor data of a subject from a glucose monitoring device;

[0097] b. predicting, using a machine learning model executed by the processor, a future occurrence of a hypoglycemic event for the subject, wherein the hypoglycemic event is associated with a predicted glucose concentration being below a predetermined threshold value for at least a predetermined time interval, wherein the machine learning model is trained based on a training method according to the present invention;

[0098] c. outputting at least one notification by using a user interface, wherein the notification comprises a diabetes management guidance message to support the subject in identifying therapeutic measures to avoid or ameliorate the occurrence of the predicted hypoglycemic event.

[0099] With respect to terms, definitions and options of the method, reference may be made to the method of training as described above or as described in further detail below.

[0100] The term “managing a predicted hypoglycemic event” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a process for avoiding or ameliorating the occurrence of the predicted hypoglycemic event.

[0101] The term “processor” as generally used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to an arbitrary logic circuitry configured for performing basic operations of a computer or system, and / or, generally, to a device which is configured for performing calculations or logic operations. In particular, the processing unit may be configured for processing basic instructions that drive the computer or system. As an example, the processor may comprise at least oneP39298

[0102] - 18 -

[0103] arithmetic logic unit (ALU), at least one floating-point unit (FPU), such as a math co-pro-cessor or a numeric coprocessor, a plurality of registers, specifically registers configured for supplying operands to the ALU and storing results of operations, and a memory, such as an LI and L2 cache memory. In particular, the processor may be a multi -core processor. Specifically, the processing unit may be or may comprise a central processing unit (CPU). Additionally or alternatively, the processor may be or may comprise a microprocessor, thus specifically the processing unit’s elements may be contained in one single integrated circuitry (IC) chip. Additionally or alternatively, the processing unit may be or may comprise one or more application-specific integrated circuits (ASICs) and / or one or more field-programmable gate arrays (FPGAs) or the like. The processing unit specifically may be configured, such as by software programming, for performing one or more evaluation operations.

[0104] The term "user interface" as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term may refer, without limitation, to an element or device which is configured for interacting with its environment, such as for the purpose of unidirectionally or bidirectionally exchanging information, such as for exchange of one or more of data or commands. For example, the user interface may be configured to share information with a user and to receive information by the user. The user interface may be a feature to interact visually with a user, such as a display, or a feature to interact acoustically with the user. The user interface, as an example, may comprise one or more of: a graphical user interface; a data interface, such as a wireless and / or a wire-bound data interface.

[0105] The term "outputting" as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term may refer, without limitation, to the process of making information available to another system, data storage, person or entity. The outputting may comprise issuing and / or triggering at least one output. As an example, the output may take place via one or more interfaces, such as a computer interface or a human-machine interface or user interface. The output, as an example, may take place in human readable format.

[0106] The term "diabetes management guidance message" as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term may refer, without limitation, to a notification relating to avoiding or ameliorate the occurrence of the predicted hypoglycemic event. For example, the diabetes management guidance message may comprise at least one recommendation such as one or more of continuing and / or starting food intake orP39298

[0107] - 19 -

[0108] drinking, informing a further person such as a health care professional, adapting insulin dosage, repeating a glucose measurement, and the like. For example, the diabetes management guidance message may comprise in addition an alert.

[0109] The term "therapeutic measures” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term may refer, without limitation, to arbitrary measures known or expected to have a positive influence for avoiding or ameliorating the occurrence of the predicted hypoglycemic event. For example, the diabetes management guidance message may comprise an overview of potential therapeutic measures, e.g. hierarchically ordered with respect to influence for avoiding or ameliorating the occurrence of the predicted hypoglycemic event.

[0110] For example, the method may comprise determining the predicted hypoglycemic event and / or during a sleep interval. The method may comprise outputting a notification by using the user interface to inform the subject of the predicted nocturnal hypoglycemic event.

[0111] In case in step b. no hypoglycemic event is predicted, the method may comprise outputting at least one notification informing that no hypoglycemic event is expected for a defined time range.

[0112] In a further aspect, a continuous glucose monitoring system is disclosed.

[0113] The continuous glucose monitoring system comprises:

[0114] at least one continuous glucose monitoring device,

[0115] a remote control configured to communicate with the at least one continuous glucose monitoring device,

[0116] one or more processors with a memory configured for storing instructions that, when executed by one or more processors, cause the one or more processors to perform the computer implemented method of training a machine learning model according to the present invention and / or the computer implemented method for managing a predicted hypoglycemic event according to the present invention.

[0117] The continuous glucose monitoring system further may comprise user interface configured for outputting at least one notification to inform the subject of the predicted hypoglycemicP39298

[0118] - 20 -

[0119] event, wherein the notification comprises a diabetes management guidance message to support the subject in identifying therapeutic measures to avoid or ameliorate the occurrence of the predicted hypoglycemic event.

[0120] With respect to terms, definitions and options of the continuous monitoring system, reference may be made to the method of training as described above or as described in further detail below.

[0121] The term "system" as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to an arbitrary set of interacting or interdependent components parts forming a whole. Specifically, the components may interact with each other in order to fulfill at least one common function. The at least two components may be handled independently or may be coupled or connectable. Consequently, the term “continuous glucose monitoring system” generally refers to a system, as defined above, configured for continuous glucose monitoring.

[0122] The term "remote control" as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a device or system of devices separate and distant from the continuous glucose monitoring device configured for controlling the continuous glucose monitoring device, e.g. execution of glucose measurements and / or data transfer. A communication between the remote control and the continuous glucose monitoring device may be a wireless data transfer. The term “wireless data transfer” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a communication link between at least two devices in a manner that the at least two devices can exchange data without the use of physical wires or cables. A wireless connection may rely on electromagnetic radiation in order to establish the communication link. Exchanging data may comprise transmitting data from a device to a further device and / or a device receiving data from a further device. The wireless connection may be or may comprise at least one of a Wireless Fidelity (Wi-Fi) connection, a Bluetooth connection, a near field connection (NFC), a Zigbee connection, a Long Range Wide Area Network (LoRaWAN) connection or the like.

[0123] The term "memory" as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special orP39298

[0124] - 21 -

[0125] customized meaning. The term specifically may refer, without limitation, to an arbitrary data storage device configured to store data. Specifically, the memory may be an electronic, magnetic and / or mechanic memory device. The memory may further be configured to store data, specifically in an organized way, such as in a database, more specifically in at least one database record.

[0126] The methods and devices as proposed herein provide a large number of advantages over methods and devices of similar kind. Specifically, the above-mentioned technical challenges may be addressed, wherein, specifically, the accuracy of predicting a hypoglycemic event may be increased. By subjecting the specific subject to predefined glycemic exposure programs, which may be carried out and / or may be monitored by a health care professional, personal glucose measurements, and optionally other physiological values, can be generated, in particular, under strictly standardized conditions. This can facilitate improved training of the machine learning model which results in an improved performance in predicting glycemic events such as hypoglycemic events. This can in turns allow to focus alarming of the patient to those situation where intervention is truly needed thus, reducing the occurrence of false alarms and alarm fatigue.

[0127] Further disclosed and proposed herein is a computer program including computer-executable instructions for performing one or both of the methods according to the present invention in one or more of the embodiments enclosed herein when the instructions are executed on a computer or computer network. Specifically, the computer program may be stored on a computer-readable data carrier and / or on a computer-readable storage medium.

[0128] As used herein, the terms “computer-readable data carrier” and “computer-readable storage medium” specifically may refer to non-transitory data storage means, such as a hardware storage medium having stored thereon computer-executable instructions. The computer-readable data carrier or storage medium specifically may be or may comprise a storage medium such as a random-access memory (RAM) and / or a read-only memory (ROM).

[0129] Thus, specifically, one, more than one or even all of method steps i. to iii, and / or a) to c), and optionally iv. to vi., as indicated above may be performed by using a computer or a computer network, preferably by using a computer program.

[0130] Further disclosed and proposed herein is a computer program product having program code means, in order to perform one or both of the methods according to the present invention inP39298

[0131] - 22 -

[0132] one or more of the embodiments enclosed herein when the program is executed on a computer or computer network. Specifically, the program code means may be stored on a computer-readable data carrier and / or on a computer-readable storage medium.

[0133] Further disclosed and proposed herein is a data carrier having a data structure stored thereon, which, after loading into a computer or computer network, such as into a working memory or main memory of the computer or computer network, may execute one or both of the methods according to one or more of the embodiments disclosed herein.

[0134] Further disclosed and proposed herein is a non-transient computer-readable medium including instructions that, when executed by one or more processors, cause the one or more processors to perform one or both of the methods according to the present invention.

[0135] Further disclosed and proposed herein is a computer program product with program code means stored on a machine-readable carrier, in order to perform one or both of the methods according to one or more of the embodiments disclosed herein, when the program is executed on a computer or computer network. As used herein, a computer program product refers to the program as a tradable product. The product may generally exist in an arbitrary format, such as in a paper format, or on a computer-readable data carrier and / or on a computer-readable storage medium. Specifically, the computer program product may be distributed over a data network.

[0136] Finally, disclosed and proposed herein is a modulated data signal which contains instructions readable by a computer system or computer network, for performing one or both of the methods according to one or more of the embodiments disclosed herein.

[0137] Referring to the computer-implemented aspects of the invention, one or more of the method steps or even all of the method steps of one or both of the methods according to one or more of the embodiments disclosed herein may be performed by using a computer or computer network. Thus, generally, any of the method steps including provision and / or manipulation of data may be performed by using a computer or computer network. Generally, these method steps may include any of the method steps, typically except for method steps requiring manual work, such as providing the samples and / or certain aspects of performing the actual measurements.

[0138] Specifically, further disclosed herein are:P39298

[0139] - 23 -

[0140] - a computer or computer network comprising at least one processor, wherein the processor is adapted to perform one or both of the methods according to one of the embodiments described in this description,

[0141] - a computer loadable data structure that is adapted to perform one or both of the methods according to one of the embodiments described in this description while the data structure is being executed on a computer,

[0142] - a computer program, wherein the computer program is adapted to perform one or both of the methods according to one of the embodiments described in this description while the program is being executed on a computer,

[0143] - a computer program comprising program means for performing one or both of the methods according to one of the embodiments described in this description while the computer program is being executed on a computer or on a computer network, - a computer program comprising program means according to the preceding embodiment, wherein the program means are stored on a storage medium readable to a computer,

[0144] - a storage medium, wherein a data structure is stored on the storage medium and wherein the data structure is adapted to perform the method according to one of the embodiments described in this description after having been loaded into a main and / or working storage of a computer or of a computer network, and

[0145] - a computer program product having program code means, wherein the program code means can be stored or are stored on a storage medium, for performing one or both of the methods according to one of the embodiments described in this description, if the program code means are executed on a computer or on a computer network.

[0146] Summarizing and without excluding further possible embodiments, the following embodiments may be envisaged:

[0147] Embodiment 1. A computer implemented method of training a machine learning model configured for predicting a hypoglycemic event for a subject, comprising:

[0148] i. receiving at least two time series of glucose sensor data, wherein each time series of glucose sensor data comprises a plurality of glucose measurements measured by a glucose monitoring device configured for detecting glucose in a bodily fluid of the subject,

[0149] wherein a first time series of the two time series of glucose sensor data is measured at least during and / or after being exposed to a first predefined program of glycemic stimuli, andP39298

[0150] - 24 -

[0151] wherein a second time series of the two time series of glucose sensor data is measured at least during and / or after being exposed to a second predefined program of glycemic stimuli, wherein the second predefined program of glycemic stimuli is different from the first predefined program of glycemic stimuli;

[0152] ii. forming a first training dataset using the first time series and forming a second training dataset using the second time series;

[0153] iii. training the machine learning model with the first training dataset and the second training dataset.

[0154] Embodiment 2. The method according to the preceding embodiment, wherein each of the predefined programs comprises at least one quantitatively and qualitatively predefined glycemic stimulus which the subject is exposed to for a predefined time period.

[0155] Embodiment 3. The method according to any one of the preceding embodiments, wherein each of the predefined programs comprises a series of quantitatively and qualitatively predefined glycemic stimuli which the subject is exposed to for a predefined time period.

[0156] Embodiment 4. The method according to any one of the preceding embodiments, wherein the first predefined program comprises at least one glycemic stimulus of a first category and the second predefined program comprises at least one glycemic stimulus of a second category, wherein the first and the second category differ, wherein the categories are selected from physical activity, food consumption, food deprivation, drug consumption, drug deprivation, insulin administration or insulin deprivation.

[0157] Embodiment 5. The method according to any one of the preceding embodiments, wherein the glycemic stimuli comprise at least one stimulus selected from the group consisting of: at least one defined meal consumption; at least one food deprivation exercise, at least one physical activity exercise; at least one insulin administration; at least one drug deprivation exercise.

[0158] Embodiment 6. The method according to any one of the preceding embodiments, wherein the first predefined program of glycemic stimuli comprises consuming a defined type and amount of carbs at defined day times, and / or wherein the second predefined program of glycemic stimuli comprises performing a defined amount of physical activity over a defined time interval and / or at defined day times.P39298

[0159] - 25 -

[0160] Embodiment 7. The method according to any one of the preceding embodiments, wherein the method comprises detecting at least one physiological value of the subject in response to the respective stimulus, wherein the physiological value is one or more of: user physical activity data from a sensor such as from a motion sensor or a heart rate sensor, glucose sensor data from the glucose monitoring device, user meal consumption data, user’s past drug administration data, drug-on-board or drug metabolite-onboard data, or a user’s disease state data.

[0161] Embodiment 8. The method according to any one of the preceding embodiments, wherein at least step i. is performed during a predetermined time interval before, during and / or after a sleep time interval, wherein the sleep time interval is defined by user entry and / or determined based on historical data collected with a sleep sensor.

[0162] Embodiment 9. The method according to any one of the preceding embodiments, wherein the machine learning model is a pretrained machine learning model, wherein the machine learning model is pretrained using historical time series of glucose measurements of one or more of: labelled data of a user population, labelled data of the subject, labelled data of at least one study, further knowledge.

[0163] Embodiment 10. The method according to the preceding embodiment, wherein the training comprises fine-tuning the pretrained machine learning model using the first training dataset and the second training dataset.

[0164] Embodiment 11. The method according to any one of the preceding embodiments, wherein the training comprises supervised learning.

[0165] Embodiment 12. The method according to any one of the preceding embodiments, wherein the first training dataset comprises the first time series of glucose sensor data and a label which indicates whether or not a hypoglycemic event occurred, wherein the second training dataset comprises the second time series of glucose sensor data and a label which indicates whether or not a hypoglycemic event occurred.

[0166] Embodiment 13. The method according to any one of the preceding embodiments, wherein the first training dataset comprises information about the exposed first predefined program of glycemic stimuli and the second training dataset comprises information about the exposed second predefined program of glycemic stimuli, and / or wherein the firstP39298

[0167] - 26 -

[0168] training dataset and the second training dataset comprise at least one determined physiological value.

[0169] Embodiment 14. The method according to any one of the preceding embodiments, wherein the machine learning model is trained using at least one classifier selected from the group consisting of: Linear Classifier, Logistic Regression, Naive Bayes Classifier, Perceptron, Support Vector Machine; Quadratic Classifier, K-Means Clustering, Boosting, Decision Tree, Random Forest; Neural network, Bayesian Network and the like.

[0170] Embodiment 15. The method according to any one of the preceding embodiments, wherein the method further comprises

[0171] iv. receiving at least one further time series of glucose sensor data, wherein each of the further time series of glucose sensor data comprises a plurality of glucose measurements measured by the at least one glucose monitoring device configured for detecting glucose in a bodily fluid of the subject, wherein the further time series is measured at least during and / or after being exposed to a further predefined program of glycemic stimuli, wherein the further predefined program of glycemic stimuli is different from the first and second predefined program of glycemic stimuli;

[0172] v. forming further training dataset using the further time series;

[0173] vi. training the machine learning model using the further training dataset.

[0174] Embodiment 16. The method according to the preceding embodiment, wherein the further predefined program of glycemic stimuli comprises administering a defined amount of insulin at defined day times.

[0175] Embodiment 17. The method according to any one of the preceding embodiments, wherein each of the time series comprise a time-ordered sequence of the glucose measurements.

[0176] Embodiment 18. The method according to any one of the preceding embodiments, wherein the glucose monitoring device is configured for continuous glucose monitoring.

[0177] Embodiment 19. A computer implemented method for managing a predicted hypoglycemic event, comprising:

[0178] a. receiving, by a processor, glucose sensor data of a subject from a glucose monitoring device;b. predicting, using a machine learning model executed by the processor, a future occurrence of a hypoglycemic event for the subject, wherein the hypoglycemic event is associated with a predicted glucose concentration being below a predetermined threshold value for at least a predetermined time interval, wherein the machine learning model is trained based on a training method according to any one of the preceding embodiments;

[0179] c. outputting at least one notification by using a user interface, wherein the notification comprises a diabetes management guidance message to support the subject in identifying therapeutic measures to avoid or ameliorate the occurrence of the predicted hypoglycemic event.

[0180] Embodiment 20. The method according to the preceding embodiment, wherein the method comprises determining the predicted hypoglycemic event before and / or during a sleep interval and outputting a notification by using the user interface to inform the subject of the predicted hypoglycemic event.

[0181] Embodiment 21. A continuous glucose monitoring system comprising:

[0182] at least one continuous glucose monitoring device,

[0183] a remote control configured to communicate with the at least one continuous glucose monitoring device,

[0184] one or more processors with a memory configured for storing instructions that, when executed by one or more processors, cause the one or more processors to perform the computer implemented method of training a machine learning model according to any one of embodiments 1 to 18 and / or the computer implemented method for managing a predicted hypoglycemic event according to any one of embodiments 19 or 20.

[0185] Embodiment 22. A computer program comprising instructions which, when the program is executed by a continuous glucose monitoring system according to embodiment 21 cause the continuous glucose monitoring system to perform the computer implemented method of training a machine learning model according to any one of embodiments 1 to 18 and / or the computer implemented method for managing a predicted hypoglycemic event according to any one of embodiments 19 or 20.

[0186] Embodiment 23. A computer-readable storage medium comprising instructions which, when the instructions are executed by a continuous glucose monitoring system according to embodiment 21 cause the continuous glucose monitoring system to perform thecomputer implemented method of training a machine learning model according to any one of embodiments 1 to 18 and / or the computer implemented method for managing a predicted hypoglycemic event according to any one of embodiments 19 or 20.

[0187] Embodiment 24. A non-transient computer-readable medium including instructions that, when executed by one or more processors, cause the one or more processors to perform the computer implemented method of training a machine learning model according to any one of embodiments 1 to 18 and / or the computer implemented method for managing a predicted hypoglycemic event according to any one of embodiments 19 or 20.

[0188] Short description of the Figures

[0189] Further optional features and embodiments will be disclosed in more detail in the subsequent description of embodiments, preferably in conjunction with the dependent claims. Therein, the respective optional features may be realized in an isolated fashion as well as in any arbitrary feasible combination, as the skilled person will realize. The scope of the invention is not restricted by the preferred embodiments. The embodiments are schematically depicted in the Figures. Therein, identical reference numbers in these Figures refer to identical or functionally comparable elements.

[0190] In the Figures:

[0191] Figure 1 shows a flowchart of an embodiment of a computer implemented method for managing a predicted hypoglycemic event; and

[0192] Figure 2 shows an embodiment of a continuous glucose monitoring system.

[0193] Detailed description of the embodiments

[0194] Figure 1 shows a flowchart of an embodiment of a computer implemented method for managing a predicted hypoglycemic event. Figure 2 shows highly schematic an embodiment of a continuous glucose monitoring system 110. The continuous glucose monitoring system 110 is configured for performing the computer implemented method for managing a predicted hypoglycemic event.The continuous glucose monitoring system 110 comprises:

[0195] at least one continuous glucose monitoring device 112,

[0196] a remote control 114 configured to communicate with the at least one continuous glucose monitoring device 112,

[0197] one or more processors 116 with a memory storing instructions that, when executed by one or more processors, cause the one or more processors to perform the computer implemented method of training a machine learning model according to the present invention and / or the computer implemented method for managing a predicted hypoglycemic event according to the present invention.

[0198] The method comprises the following method steps, which, as an example, may be performed in the given order. However, a different order is also feasible. Further, it is possible to perform two or more of the method steps simultaneously or in a fashion overlapping in time. Further, it is also possible to perform one, more than one or even all of the method steps repeatedly.

[0199] The method comprises:

[0200] a. (118) receiving, by the processor 116, glucose sensor data of a subject from a glucose monitoring device 112;

[0201] b. (120) predicting, using a machine learning model executed by the processor 116, a future occurrence of a hypoglycemic event for the subject, wherein the hypoglycemic event is associated with a predicted glucose concentration being below a predetermined threshold value for at least a predetermined time interval, wherein the machine learning model is trained based on a training method according to the present invention;

[0202] c. (122) outputting at least one notification by using a user interface 124, e.g. to inform the subject of the predicted hypoglycemic event, wherein the notification comprises a diabetes management guidance message to support the subject in identifying therapeutic measures to avoid or ameliorate the occurrence of the predicted hypoglycemic event.

[0203] In case in step b. 120 no hypoglycemic event is predicted, the method may comprise outputting at least one notification informing that no hypoglycemic event is expected for a defined time range.

[0204] The training method comprises the following steps:i. (126) receiving at least two time series of glucose sensor data, wherein each time series of glucose sensor data comprises a plurality of glucose measurements measured by the glucose monitoring device 112 configured for detecting glucose in a bodily fluid of the subject, wherein a first time series of the two time series of glucose sensor data is measured at least during and / or after being exposed to a first predefined program of glycemic stimuli, and wherein a second time series of the two time series of glucose sensor data is measured at least during and / or after being exposed to a second predefined program of glycemic stimuli, wherein the second predefined program of glycemic stimuli is different from the first predefined program of glycemic stimuli;

[0205] ii. (128) forming a first training dataset using the first time series and forming a second training dataset using the second time series;

[0206] iii. (130) training the machine learning model with the first training dataset and the second training dataset.

[0207] The prediction of a hypoglycemic event may be a qualitative prediction, i.e. expected occurrence of a hypoglycemic event or expected non-occurrence of a hypoglycemic event. Additionally or alternatively, the prediction may comprise a quantitative prediction, e.g. indication likelihood of occurrence of a hypoglycemic event, e.g. in percent.

[0208] The machine learning model may be a pretrained model. Thus, the training of the machine learning model may comprise fine-tuning of the machine learning model. For example, the machine learning model may be pretrained using historical time series of glucose measurements of one or more of: labelled data of a user population, labelled data of the subject, labelled data of at least one study, further knowledge. The pretraining of the machine learning model may comprise using at least one classifier selected from the group consisting of: Linear Classifier, Logistic Regression, Naive Bayes Classifier, Perceptron, Support Vector Machine; Quadratic Classifier, K-Means Clustering, Boosting, Decision Tree, Random Forest; Neural network, Bayesian Network and the like. For example, the pretrained model may be generated as described in Jensen et al. “Prediction of Nocturnal Hypoglycemia From Continuous Glucose Monitoring Data in People With Type 1 Diabetes: A Proof-of-Concept Study”, Journal of Diabetes Science and Technology 2020, Vol. 14(2) 250- 256.

[0209] The glucose monitoring device may comprise at least one sensor configured for qualitatively or quantitatively detecting the presence and / or the concentration of glucose in bodily fluid. For example, the bodily fluid may be selected from the group consisting of blood and interstitial fluid. However, additionally or alternatively, one or more other types of bodily fluidsP39298

[0210] - 31 -

[0211] may be used, such as saliva, tear fluid, urine or other body fluids. The glucose monitoring device may be configured for being at least partially inserted into a body tissue of the user. The glucose monitoring device may be configured for continuous glucose monitoring (CGM). The glucose sensor data may be or may comprise data obtained by using the glucose monitoring device 112, e.g. a raw sensor signal, a processed sensor signal and / or information derived from the sensor signal. For example, the glucose sensor data may comprise a glucose concentration value, e.g. a molar concentration value or a mass concentration value. The time series of glucose sensor data may be or may comprise a sequence of glucose measurements, which are ordered with respect to measurement time, e.g. starting with the oldest glucose measurement. Each of the time series may comprise a time-ordered sequence of the glucose measurements. Each of the glucose measurements may comprise a time stamp.

[0212] The predefined program of glycemic stimuli may be or may comprise a program comprising at least one predefined step of exposing the subject to at least one glycemic stimulus. For example, the predefined program of glycemic stimuli may be standardized, e.g. in the sense that for each subject or a subject of a specific category, e.g. age, disease state and the like, an identical predefined program of glycemic stimuli is used. Alternatively, the predefined program of glycemic stimuli may be adapted for each subject. For example, the predefined program of glycemic stimuli may be defined by a health care professional. The exposure to the glycemic stimuli may be performed under monitoring and / or control of a health care professional.

[0213] The predefined glycemic stimuli exposure program may comprise at least one arbitrary stimulus expected to result in a change in glucose level and / or expected to have an influence on the subject’s glucose level. For example, each of the predefined programs comprises at least one quantitatively and qualitatively predefined glycemic stimulus which the subject is exposed to for a predefined time period. For example, each of the predefined programs comprises a series of quantitatively and qualitatively predefined glycemic stimuli which the subject is exposed to for a predefined time period. For example, the glycemic stimuli comprise at least one stimulus selected from the group consisting of at least one defined meal consumption; at least one food deprivation exercise, at least one physical activity exercise; at least one insulin administration; at least one drug deprivation exercise.

[0214] The first predefined program may comprise at least one glycemic stimulus of a first category and the second predefined program may comprise at least one glycemic stimulus of a second category. The first and the second category may differ. For example, the categories are selected from physical activity, food consumption, food deprivation, drug consumption, drugP39298

[0215] - 32 -

[0216] deprivation, insulin administration or insulin deprivation. For example, the first predefined program of glycemic stimuli comprises consuming a defined type and amount of carbs at defined day times. For example, the second predefined program of glycemic stimuli comprises performing a defined amount of physical activity over a defined time interval and / or at defined day times.

[0217] For example, the method may comprise receiving at least one glucose measurement, in particular a first time series of glucose sensor data, from the glucose monitoring device before the subject being exposed to a first and / or second predefined program of glycemic stimuli. These data may be used for comparison with glucose sensor data received during and / or after the subject being exposed to a first or second predefined program of glycemic stimuli.

[0218] The method may comprise detecting, in addition to the time series of glucose sensor data measured during and / or after the exposure to the predefined glycemic stimuli, at least one physiological value of the subject in response to the respective stimulus, e.g. physiological data that reflect the subject’s behavior in response to the predefined glycemic stimulus exposure. Said physiological value may be recorded during and / or after the execution of the first and / or second predefined program of glycemic stimuli. The physiological value of the subject may be one or more of user physical activity data from a sensor such as from a motion sensor or a heart rate sensor, glucose sensor data from the glucose monitoring device, user meal consumption data, user’s past drug administration data, drug-on-board or drug metabolite-on-board data, or a user’s disease state data. The user may be the subject. The physiological value of the subject may be used as additional input fortraining of the machine learning model.

[0219] The first time series of glucose sensor data may be used as data for the first training dataset. The second time series of glucose sensor data may be used as data for the second training dataset. The first training dataset may comprise information about the first predefined program of glycemic stimuli and / or the physiological value of the subject in response to the respective stimulus. The second training dataset may comprise information about the second predefined program of glycemic stimuli and / or the physiological value of the subject in response to the respective stimulus. The first and / or second training dataset may comprise additional information on the subject such as one or more of age, health data, activity data, and the like. The forming of the training dataset may comprise bringing the received data in a data format which can be used for training the machine learning model. The forming of the training dataset may comprise at least one preprocessing step, e.g. comprising smoothing, filtering and the like.P39298

[0220] - 33 -

[0221] As outlined above, step iii. 130 comprises training the machine learning model with the first training dataset and the second training dataset. The training with the first training dataset and the second training dataset may be performed at the same time, e.g. in parallel, or subsequently. For example, the machine learning model may be firstly trained on the first training dataset and, subsequently, may be trained on the second training dataset. In this case, the method may comprise receiving, in step i. 126, the first time series only, forming the first training dataset in step ii. 128 and training the machine learning model on the first training dataset in step iii.. Subsequently, steps i. 126 to iii. 130 may be performed for the second time series. Other embodiments and orders of steps may be possible.

[0222] The training of the machine learning model may comprise fine-tuning the pretrained machine model using the first training dataset and the second training dataset. The additional training of the pretrained machine learning model, such as a population user data trained model, can allow personalizing the pretrained machine learning model for the specific subject. This can be possible by training using the recorded specific user data which was collected during the predefined programs of glycemic stimuli.

[0223] The machine learning model may be trained using at least one classifier selected from the group consisting of Linear Classifier, Logistic Regression, Naive Bayes Classifier, Perceptron, Support Vector Machine; Quadratic Classifier, K-Means Clustering, Boosting, Decision Tree, Random Forest; Neural network, Bayesian Network and the like.

[0224] For example, the training, in particular the fine-tuning, comprises supervised learning, in particular learning based on labeled training data. The labeled training data may comprise information about the glycemic stimuli the subject was exposed to. The labeled training data may comprise information about one or more of information about a glucose level development within a predefined time range during and / or after exposing to the glycemic stimuli, information about occurrence of a hypoglycemic event within a predefined time range during and / or after exposing to the glycemic stimuli. For example, the first training dataset may comprise the first time series of glucose sensor data and a label which indicates whether or not a hypoglycemic event occurred. For example, the second training dataset may comprise the second time series of glucose sensor data and a label, which indicates whether or not a hypoglycemic event occurred. For example, the first training dataset may comprise information about the exposed first predefined program of glycemic stimuli and the second train-P39298

[0225] - 34 -

[0226] ing dataset may comprise information about the exposed second predefined program of glycemic stimuli. Additionally or alternatively, the first training dataset and the second training dataset may comprise at least one determined physiological value.

[0227] The training method specifically may comprise splitting the training dataset into a training dataset and a test dataset. Thus, more specifically, the splitting into a training dataset and a test dataset may comprise x%-y% splitting of the training dataset, with x% being in the range 60 to 90, specifically in the range 65 to 75, and more specifically x%=70, and with y%=100-x. Therein, x% denotes the training dataset and y% denotes the test dataset.

[0228] Steps ii. 128 and iii. 130 may be carried out during and / or after data recording in step i..

[0229] The method may be used for predicting a nocturnal hypoglycemic event, in particular for predicting a hypoglycemic event occurring during a time the subject is sleeping. At least step i. may be performed during a predetermined time interval before, during and / or after a sleep time interval. The sleep time interval may be defined by user entry, e.g. via at least one user interface, and / or determined based on historical data collected with a sleep sensor, such as a sleep sensor configured for polysomnography or actigraphy. The training in step iii. may be carried out at defined time points, e.g. before sleep. The physiological value may be collected before and / or during and / or after exposing the subject to the glycemic stimuli including the time period during sleep. The method may further comprise at least one step of determining a medical risk associated with the predicted hypoglycemic event, e.g. during the sleep time interval. The method may further comprise outputting a notification to a display device to inform the subject and / or a further person, e.g. a health care professional, of the predicted hypoglycemic event if the medical risk during the sleep period is above a reference threshold.

[0230] The method may comprise receiving a first time series of glucose sensor data from the glucose monitoring device before and / or during and / or after being exposed to the first predefined program of glycemic stimuli. For example, the first predefined program of glycemic stimuli may comprise consuming a defined type and amount of carbs at defined time points. The glucose measurement values may be recorded before and / or until and / or after the consumption and may constitute the first time series of glucose sensor data. The method may comprise training the machine learning model to associate the first time series of glucose sensor data with the first predefined program of glycemic stimuli.

[0231] The method may comprise receiving a second time series of glucose sensor data from the glucose monitoring device before and / or during and / or after being exposed to the secondP39298

[0232] - 35 -

[0233] predefined program of glycemic stimuli. For example, the second predefined program of glycemic stimuli may comprise performing a defined amount of physical activity over a defined time interval. The glucose measurement values may be recorded before and / or until and / or after the activity constitute the second time series of glucose sensor data. The method may comprise training the machine learning model to associate the second time series of glucose sensor data with the second predefined program of glycemic stimuli.

[0234] The method may further comprises

[0235] iv. receiving at least one further time series of glucose sensor data, wherein each of the further time series of glucose sensor data comprises a plurality of glucose measurements measured by the at least one glucose monitoring device 112 configured for detecting glucose in a bodily fluid of the subject, wherein the further time series is measured at least during and / or after being exposed to a further predefined program of glycemic stimuli, wherein the further predefined program of glycemic stimuli is different from the first and second predefined program of glycemic stimuli; v. forming further training dataset using the further time series;

[0236] vi. training the machine learning model using the further training dataset.

[0237] The method may comprise receiving a third time series of glucose sensor data from the glucose monitoring device 112 before and / or during and / or after being exposed to the third predefined program of glycemic stimuli. For example, the third predefined program of glycemic stimuli may comprise administering a defined amount of insulin at defined time points. The glucose measurement values may be recorded before and / or until and / or after the activity constitute the third time series of glucose sensor data. The method may comprise training the machine learning model to associate the third time series of glucose sensor data with the third predefined program of glycemic stimuli.

[0238] In step c. 122, the diabetes management guidance message may be or may comprise a notification relating to avoiding or ameliorate the occurrence of the predicted hypoglycemic event. For example, the diabetes management guidance message may comprise at least one recommendation such as one or more of continuing and / or starting food intake or drinking, informing a further person such as a health care professional, adapting insulin dosage, repeating a glucose measurement, and the like. For example, the diabetes management guidance message may comprise in addition an alert.P39298

[0239] - 36 -

[0240] The therapeutic measures may be at least one arbitrary measure known or expected to have a positive influence for avoiding or ameliorating the occurrence of the predicted hypoglycemic event. For example, the diabetes management guidance message may comprise an overview of potential therapeutic measures, e.g. hierarchically ordered with respect to influence for avoiding or ameliorating the occurrence of the predicted hypoglycemic event.

[0241] For example, the method may comprise determining the predicted hypoglycemic event before and / or during a sleep interval and outputting a notification by using the user interface to inform the subject of the predicted hypoglycemic event.

[0242]

[0243] stimuli:

[0244] The following predefined programs of glycemic stimuli are meant to illustrate rather than limit the invention and can be carried out alone or combined with each other to train the a machine learning model. Any of the below predefined programs of glycemic stimuli can be used for the first or second predefined program of glycemic stimuli. The first and second time series of glucose sensor data are being measured at least during and / or after being exposed to the respective predefined program of glycemic stimuli, optionally also before being exposed to the respective predefined program of glycemic stimuli.

[0245] 1) Predefined program of food consumption stimuli

[0246] The subject is asked to consume a predefined amount of food characterized by a defined amount of carbohydrates. Carbohydrates can be broken down depending on how fast they affect the blood glucose level into fast-acting, medium-acting and slow-acting. As a result, consumption of fast acting carbohydrates increases the blood glucose level much faster than slow acting carbohydrates. Fast-acting carbohydrates comprise, for example, glucose tablets, glucose drinks, full-sugar soft drinks or squashes, jellies (not diet), and sweets. Medium -acting carbohydrates comprise, for example, bread, pasta, potatoes, yams, breakfast cereal, and couscous. Slow-acting carbohydrates comprise, for example, pearl barley, peas, beans, lentils, sweetcorn, pumpkin. By training the machine learning model with a predefined program of food consumption of a defined amount and type of carbohydrates the model can be trained to accurately predict the glucose level of the subject.

[0247] To this end the subject may be asked to consume the following predefined food stimulus, to docket the food consumption with the computer implemented method of training a machine learning model 5 minutes before the food consumption and optionally to refrain from otherbehaviors that may significantly influence the blood glucose level such as insulin administration, physical activity, drug consumption, and consumption additional food, for a period of Ih or 2h following the consumption of the predefined food stimulus. The predefined food stimulus may be

[0248] o 20g, 40g, 60g or 80g of fast acting carbohydrates (such as a glucose drink) o 20g, 40g, 60g or 80g of medium acting carbohydrates (such as a breakfast cereal meal)

[0249] o 20g, 40g, 60g or 80g of slow acting carbohydrates (such as pumpkin soup meal)

[0250] 2) Predefined program of food deprivation stimuli

[0251] To this end the subject may be asked to refrain from consuming food after finishing the previous meal (e.g. breakfast) for a period of 3h, 4h or 5h and optionally to also refrain from other behaviors that may significantly influence the blood glucose level such as insulin administration, physical activity, and drug consumption, for the above period.

[0252] 3) Predefined program of insulin administration stimuli

[0253] There are different insulin types that differ in regard to how rapidly an how long the insulin takes effect on the blood glucose level, examples are rapid-acting insulin (e.g. Lyspro), regular human insulin, intermediate-acting insulin (e.g. neutral protamine hagedom (NPH) insulin), and long-acting insulin (e.g. Detemir or Lantus). Accordingly, a predefined insulin administration stimuli program may take into account the type of insulin used for the training, e.g. by limiting the training to the use of one type of insulin or by docketing the insulin type used with the computer implemented method of training a machine learning model. Since the injection site of the insulin such as abdomen vs. arm may also have an impact on the time it takes for the insulin to take effect on the blood glucose level it may be considered to refrain from changing the injection site during the training program.

[0254] The subject may be asked to administer the following predefined insulin bolus amount, to docket the administered insulin bolus amount with the computer implemented method of training a machine learning model 5 minutes before the administration and optionally to refrain from other behaviors that may significantly influence the blood glucose level such as additional insulin administration, physical activity, drug consumption, and food consumption, for a period of between 30 min to 4h, such as 2h following the predefined insulin administration. In another embodiment the administration of the predefined insulin bolus amount is carried out when the subject has not received any insulin administration for a period of 30 min to 2h prior administration of the predefined insulin bolus amount. In anotherembodiment the administration of the predefined insulin bolus amount is carried out when the last or the last two blood glucose measurements carried out within the last 30 minutes were within a euglycemic range of 70 to 180 mg / dl.

[0255] The predefined insulin bolus amount may be:

[0256] 5U, 10U, 20U, 3 OU, or 40U of a defined insulin.

[0257] Considering that overdosing of insulin is associated with a risk of creating a hypoglycemia which is associated with a significant health risk, the predefined insulin administration stimuli program should be carried out under the guidance of a health care professional.

[0258] 4) Predefined program of insulin deprivation stimuli

[0259] The subject may be asked to refrain from administering insulin and optionally to refrain from other behaviors that may significantly influence the blood glucose level such as additional insulin administration, physical activity, drug consumption, and food consumption, for a period of between 30 min to 4h, such as 2h following the last prior insulin administration.

[0260] Considering that underdosing of insulin is associated with a risk of creating a hyperglycemia which is associated with a significant health risk, the predefined insulin deprivation stimuli program should be carried out under the guidance of a health care professional.

[0261] 5) Predefined program of physical activity exercise stimuli

[0262] Physical activity has a blood glucose level lowering effect. Different types of physical activity exercises can have a different impact on the blood glucose levels, depending on the intensity of the physical activity. A low-intensity physical activity exercise uses a larger percentage of energy from fat, as the body does not need to produce energy quickly and efficiently to maintain the activity. During a low intensity, such as walk a mile at normal walking speed, only 20% of the energy will be supplied in the form of glucose and 80% from fat. On the other hand, higher-intensity physical activity exercise gets a larger percentage of its energy from carbohydrates, as the body is able to produce this quickly, making glucose the main source of energy in such activities. For example, when running a mile at a strenuous pace as much as 80% of the energy may come from glucose. The individual body parameters such as body mass index, size, disease state and fitness of a subject have an impact on what constitutes a low, moderate or vigorous intensity physical activity exercise for this subject. The level of physical activity intensity can for example be determined based on the subject’s percentage of maximum heart rate, a measure generally known in the art: e.g.low-intensity physical activity exercise: up to 49% of the subject’s maximum heart rate

[0263] moderate-intensity physical activity exercise: 50 to 69% of the subject’s maximum heart rate

[0264] - vigurous-intensity physical activity exercise: 70 to 85% of the subject’s maximum heart rate.

[0265] To train the computer implemented method of training a machine learning model according to the invention the subject is tasked to carry out the following predefined physical exercise program. The subject may be asked to docket the predefined physical exercise program with the computer implemented method of training a machine learning model 5 minutes before the exercise program is initiated and to optionally to refrain from other behaviors that may significantly influence the blood glucose level such as insulin administration, drug consumption, and food consumption, for a period of Ih or 2h following the predefined physical exercise program. The predefined food stimulus may be

[0266] Conduct a low-intensity physical activity exercise (e.g. at an average of 40% of the subject’s maximum heart rate while essentially remaining within the limits of up to 49% of the subject’s maximum heart rate) for 10 min to 2h, e.g. for 30 minutes Conduct a moderate-intensity physical activity exercise (e.g. at an average of 60% of the subject’s maximum heart rate while essentially remaining within the limits of 50 to 69% of the subject’s maximum heart rate) for 10 min to 2h, e.g. for 30 minutes Conduct a vigorous-intensity physical activity exercise (e.g. at an average of 80% of the subject’s maximum heart rate while essentially remaining within the limits of 70 to 85% of the subject’s maximum heart rate) for 10 min to 2h, e.g. for 30 minutes

[0267] Optionally the step of drug docketing the predefined physical exercise stimuli program with the computer implemented method of training a machine learning model involves the docketing of the level of physical intensity of the predefined physical exercise program (e.g. low, moderate of vigorous exercise) and the calculated energy consumed during the program (e.g. carbohydrate amount) which can be readily determined using conventional methods.

[0268] Considering that intense physical activity is associated with a risk of creating a hypoglycemia which is associated with a significant health risk, the predefined physical exercise stimuli program should be carried out under the guidance of a health care professional.

[0269] 6) Predefined program of drug consumption stimuliCertain drugs are known to influence the blood glucose level in a subject: corticosteroids, niacin, certain antipsychotic drugs, certain decongestants such as Sudafed, statins and beta blockers may increase the blood glucose level. Metformin and GLP-1 receptor agonists, in turn lowers the blood glucose level.

[0270] Accordingly, a predefined program of drug consumption stimuli may take into account the administration of a blood glucose influencing drug for the training, e.g. by docketing the administration of the blood glucose influencing drug used with the computer implemented method of training a machine learning model.

[0271] The subject may be asked to administer the typical dose of the blood glucose influencing drug, and to docket the administered with the computer implemented method of training a machine learning model 5 minutes before the administration and optionally to refrain from other behaviors that may significantly influence the blood glucose level such as insulin administration, physical activity, drug consumption, and food consumption, for a period of between 30 min to 4h, such as 2h following the predefined drug administration.P39298

[0272] List of reference numbers

[0273] 110 continuous glucose monitoring system 112 continuous glucose monitoring device 114 remote control

[0274] 116 processor

[0275] 118 step a.

[0276] 120 step b.

[0277] 122 step c.

[0278] 124 user interface

[0279] 126 step i.

[0280] 128 step ii.

[0281] 130 step iii.

Claims

P39298- 42 -November 28, 2925Claims1. A computer implemented method of training a machine learning model configured for predicting a hypoglycemic event for a subject, comprising:i. (126) receiving at least two time series of glucose sensor data, wherein each time series of glucose sensor data comprises a plurality of glucose measurements measured by a glucose monitoring device (112) configured for detecting glucose in a bodily fluid of the subject,wherein a first time series of the two time series of glucose sensor data is measured at least during and / or after being exposed to a first predefined program of glycemic stimuli, andwherein a second time series of the two time series of glucose sensor data is measured at least during and / or after being exposed to a second predefined program of glycemic stimuli, wherein the second predefined program of glycemic stimuli is different from the first predefined program of glycemic stimuli;ii. (128) forming a first training dataset using the first time series and forming a second training dataset using the second time series;iii. (130) training the machine learning model with the first training dataset and the second training dataset,iv. wherein the first predefined program comprises at least one glycemic stimulus of a first category and the second predefined program comprises at least one glycemic stimulus of a second category, wherein the first and the second category differ, wherein the categories are selected from physical activity, food consumption, food deprivation, drug consumption, drug deprivation, insulin administration or insulin deprivation, wherein the glycemic stimuli comprise at least one stimulus selected from the group consisting of: at least one defined meal consumption; at least one food deprivation exercise, at least one physical activity exercise; at least one insulin administration; at least one drug deprivation exercise.v. receiving at least one further time series of glucose sensor data, wherein each of the further time series of glucose sensor data comprises a plurality of glucose measurements measured by the at least one glucose monitoring device (112) configured for detecting glucose in a bodily fluid of the subject, wherein the further time series is measured at least during and / or after being exposed to a further predefined programP39298- 43 -of glycemic stimuli, wherein the further predefined program of glycemic stimuli is different from the first and second predefined program of glycemic stimuli; vi. forming further training dataset using the further time series;vii. training the machine learning model using the further training dataset.

2. The method according to the preceding claim, wherein each of the predefined programs comprises at least one quantitatively and qualitatively predefined glycemic stimulus which the subject is exposed to for a predefined time period, or wherein each of the predefined programs comprises a series of quantitatively and qualitatively predefined glycemic stimuli which the subject is exposed to for a predefined time period.

3. The method according to any one of the preceding claims, wherein the method comprises detecting at least one physiological value of the subject in response to the respective stimulus, wherein the physiological value is one or more of: user physical activity data from a sensor such as from a motion sensor or a heart rate sensor, glucose sensor data from the glucose monitoring device, user meal consumption data, user’s past drug administration data, drug-on-board or drug metabolite-on-board data, or a user’s disease state data.

4. The method according to any one of the preceding claims, wherein at least step i. (126) is performed during a predetermined time interval before, during and / or after a sleep time interval, wherein the sleep time interval is defined by user entry and / or determined based on historical data collected with a sleep sensor.

5. The method according to any one of the preceding claims, wherein the machine learning model is a pretrained machine learning model, wherein the machine learning model is pretrained using historical time series of glucose measurements of one or more of: labelled data of a user population, labelled data of the subject, labelled data of at least one study, further knowledge.

6. The method according to any one of the preceding claims, wherein the first training dataset comprises information about the exposed first predefined program of glycemic stimuli and the second training dataset comprises information about the exposed second predefined program of glycemic stimuli, and / or wherein the first training dataset and the second training dataset comprise at least one determined physiological value.P39298- 44 -7. The method according to any one of the preceding claims, wherein the machine learning model is trained using at least one classifier selected from the group consisting of: Linear Classifier, Logistic Regression, Naive Bayes Classifier, Perceptron, Support Vector Machine; Quadratic Classifier, K-Means Clustering, Boosting, Decision Tree, Random Forest; Neural network, Bayesian Network and the like.viii.

8. A computer implemented method for managing a predicted hypoglycemic event, comprising:a. (118) receiving, by a processor (116), glucose sensor data of a subject from a glucose monitoring device (112);b. (120) predicting, using a machine learning model executed by the processor (116), a future occurrence of a hypoglycemic event for the subject, wherein the hypoglycemic event is associated with a predicted glucose concentration being below a predetermined threshold value for at least a predetermined time interval, wherein the machine learning model is trained based on a training method according to any one of the preceding claims;c. (122) outputting at least one notification by using a user interface (124), wherein the notification comprises a diabetes management guidance message to support the subject in identifying therapeutic measures to avoid or ameliorate the occurrence of the predicted hypoglycemic event.

9. The method according to the preceding claim, wherein the method comprises determining the predicted hypoglycemic event before and / or during a sleep interval.

10. A continuous glucose monitoring system (110) comprising:at least one continuous glucose monitoring device (112),a remote control (114) configured to communicate with the at least one continuous glucose monitoring device (112),one or more processors (116) with a memory configured for storing instructions that, when executed by one or more processors (116), cause the one or more processors (116) to perform the computer implemented method of training a machine learning model according to any one of claims 1 to 9 and / or the computer implemented method for managing a predicted hypoglycemic event according to any one of claims 10 or 11.P39298- 45 -11. A computer program comprising instructions which, when the program is executed by a continuous glucose monitoring system (110) according to claim 12 cause the continuous glucose monitoring system (110) to perform the computer implemented method of training a machine learning model according to any one of claims 1 to 9 and / or the computer implemented method for managing a predicted hypoglycemic event according to any one of claims 10 or 11.

12. A computer-readable storage medium comprising instructions which, when the instructions are executed by a continuous glucose monitoring system (110) according to claim 12 cause the continuous glucose monitoring system (110) to perform the computer implemented method of training a machine learning model according to any one of claims 1 to 9 and / or the computer implemented method for managing a predicted hypoglycemic event according to any one of claims 10 or 11.

13. A non-transient computer-readable medium including instructions that, when executed by one or more processors, cause the one or more processors to perform the computer implemented method of training a machine learning model according to any one of claims 1 to 9 and / or the computer implemented method for managing a predicted hypoglycemic event according to any one of claims 10 or 11.

Images