Treatment decision support for treatment of type 2 diabetes mellitus

Abstract

The present disclosure relates to a system for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus, wherein the system comprises: a processing unit; a memory; an interface; an input device; wherein, the memory comprises instructions that, when executed by the processing unit, carries out a method for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus.

Description

[0001] Treatment decision support for treatment of type 2 diabetes mellitus

[0002] The present invention relates to systems and methods for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus.

[0003] Background

[0004] Type 2 diabetes, characterized by elevated blood glucose levels, is a prevalent form of diabetes. This condition arises when the body fails to utilize or later sufficiently produce insulin, a hormone responsible for facilitating glucose absorption into cells as an energy source. Consequently, excessive glucose accumulates in the bloodstream, impeding its delivery to cells and resulting in complications affecting the circulatory, nervous, and immune systems.

[0005] Although formerly referred to as adult-onset diabetes, both type 1 and type 2 diabetes can manifest during childhood and adulthood. While type 2 diabetes is more commonly observed in older individuals, the recent rise in childhood obesity has led to an increased incidence of type 2 diabetes in younger populations.

[0006] While there is currently no permanent cure for type 2 diabetes, the disease is typically managed and sometimes temporarily reversed by weight loss, proper nutrition, and regular exercise. However, if these lifestyle modifications fail to sufficiently regulate blood sugar levels, medical interventions such as diabetes medications or insulin therapy may be prescribed. Presently, pharmaceutical treatments for type 2 diabetes encompass various options, each carrying its own advantages, risks, and side effects.

[0007] Existing guidelines and protocols aim to guide treatment decisions for patients.

[0008] However, these recommendations often provide generalized approaches rather than personalized solutions. Further, in practice, none of these approaches guides primary care physicians in a clear, interactive, and transparent manner to make decisions suited for the individual patient.

[0009] Consequently, there is a strong unmet need for treatment decision support systems that can deliver accurate and personalized medical treatment options for the management of type 2 diabetes. Specifically, a need for a comprehensive and tailored decision support system for healthcare professionals in managing this chronic condition. Summary

[0010] The systems and methods for treatment decision support of the prior art does not take into account in an efficient way crucial factors such as the patient's specific circumstances, laboratory results, coexisting medical conditions, and potential interactions with other medications, despite their significance in tailoring individualized treatments.

[0011] In view of this, a purpose of the present invention is to provide a novel and efficient approach to meet these needs.

[0012] Therefore, in a first aspect, the present disclosure relates to a system for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus, wherein the system comprises:

[0013] • a processing unit;

[0014] • a memory;

[0015] • an interface;

[0016] • an input device; wherein, the memory comprises instructions that, when executed by the processing unit, carries out a method comprising:

[0017] I. performing an initial assessment of the treatment options, based on patient data, wherein the initial assessment comprises: i. selecting treatment options, comprising excluding disadvantageous treatment options, such as treatment options that are contradicted by one or more parameters in the patient data; ii. providing an initial scoring to the selected treatment options, wherein the initial scoring is based on expected effects of the selected treatment options for the patient, such as based on treatment guidelines;

[0018] II. obtaining, by the input device, continuous glucose monitoring (CGM) data of the patient;

[0019] III. extracting a plurality of CGM features from the CGM data; IV. predicting, by a machine learning model that is trained to predict outcome CGM features based on the CGM features, the outcome of each selected treatment option;

[0020] V. scoring the selected treatment options, including a CGM scoring, and a price scoring;

[0021] VI. ranking the selected treatment options based on the initial scoring, the CGM scoring, and the price scoring;

[0022] VII. providing, by the interface, at least one recommended treatment option, wherein the recommended treatment options exclude combinations of selected treatment options that are expected to have harmful drug interaction; thereby providing the treatment decision support of the patient for the plurality of treatment options.

[0023] The system efficiently allows for a determination on an optimal personalized treatment decision support of a patient, based on a plurality of treatment options for treatment of type 2 diabetes mellitus. In particular, the system is arranged to take into account measured CGM data from the patient, by an input device, for example a CGM device that is arranged to measure the blood glucose values of the patient. This allows for a more accurate prediction of the optimal personalized treatment.

[0024] Further, the system is arranged to combine the CGM data with patient data and current guidelines of diabetes treatment to predict, by machine learning models, individual predicted outcomes of CGM features. By having arranged the system so that each step it carries out is presented through an interface to the user, typically the primary care physician, the process by which an optimal treatment is identified and recommended to the patient is made transparent.

[0025] The system is typically arranged to carry out a method that includes a guideline module, a CGM module and a prioritization module, so that, when combined as disclosed herein results in the efficient and accurate recommendation of a treatment option for the patient.

[0026] Typically, the initial assessment module comprises excluding, selecting, and / or providing an initial scoring to, a number of treatment options, based on patient data. The method may for example comprise excluding treatment options that is expected to result in a contraindication for the patient. The method may comprise a step of providing an initial scoring comprising providing, for each non-excluded treatment option, a score based on patient specific guidelines for treatment of type 2 diabetes mellitus.

[0027] The patient data may for example comprise patient demographics, clinical data, previous or present health conditions, year of diabetes onset, and / or a target value for HbA1c. The clinical data may for example comprise one or more measurements of markers for metabolic control (such as HbA1c, Fasting glucose, and / or C-peptide), kidney function (such as creatinine, glomerular filtration rate, and / or albuminuria) and liver function (such as ASAT, ALAT, PK, Albumin, and / or bilirubin).

[0028] In the present example, the initial assessment module provides an output to the CGM module which is arranged to apply points to the selected treatment options, based on CGM features and the patient data. The CGM features may have been identified from CGM data of the patient that has been obtained continuously for a period of time, such as a number of recent days, for example the last seven days.

[0029] The CGM features are typically features that are indicative of the status of the type 2 diabetes mellitus of the patient, but may for example comprise the meal features, time in one or more glycemic ranges, hypoglycemic features, mean glucose features, hyperglycemic features, and / or fluctuation features.

[0030] The method typically comprises a step of predicting by a trained machine learning model, the outcome of each selected treatment option. Thus, the method may rely on multiple machine learning models, and wherein the result of each selected treatment option is predicted by a different machine learning model.

[0031] Further, the method typically comprises a step of scoring, comprising, for each selected treatment option, providing an initial score, a CGM score and / or a price score, wherein a better initial score means a better outcome for the patient according to the treatment guidelines, CGM scoring means a better outcome for the patient according to the machine learning model(s), and similarly a better pricing score means a lower price. Based on the scoring of the selected treatment options, the method is arranged to provide a recommendation of the best selected treatment option that is not expected to have any harmful drug interactions.

[0032] In a second aspect, the present disclosure relates to a computer-implemented method for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus, the method comprising:

[0033] • performing an initial assessment of the treatment options, based on patient data, wherein the initial assessment comprises: i. selecting treatment options, comprising excluding disadvantageous treatment options, such as treatment options that are contradicted by one or more parameters in the patient data; ii. providing an initial scoring to the selected treatment options, wherein the initial scoring is based on expected effects of the selected treatment options for the patient, such as based on treatment guidelines;

[0034] • obtaining continuous glucose monitoring (CGM) data of the patient, comprising glucose values of the patient obtained during a predetermined time period;

[0035] • extracting a plurality of CGM features from the CGM data;

[0036] • predicting, for each selected treatment option, by a trained machine learning model, the outcome of each selected treatment option;

[0037] • scoring the selected treatment options, including a CGM scoring, and a price scoring;

[0038] • ranking the selected treatment options based on the initial scoring, the CGM scoring, and the price scoring;

[0039] • providing at least one recommended treatment option, wherein the recommended treatment options exclude combination of selected treatment options that are expected to have harmful drug interaction; thereby providing the treatment decision support of the patient for the plurality of treatment options. In a further aspect, the present disclosure relates to a computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the steps of the computer-implemented method for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus as disclosed elsewhere herein.

[0040] In yet a further aspect, the present disclosure relates to a system for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus, wherein the system is arranged to carry out the computer-implemented method for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus as disclosed elsewhere herein.

[0041] In even yet a further aspect, the present disclosure relates to a method of training a machine learning model for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus, the method comprising:

[0042] • obtaining patient data of a plurality of patients and CGM data of the patients before starting a new treatment and after starting said treatment, the CGM data comprising: i. pretreatment CGM data, comprising CGM data before the start of the treatment; ii. treatment CGM data, comprising CGM data after the start of the treatment, such as when an effective dose of the treatment has been reached in the patient;

[0043] • extracting, for each patient, a plurality of pretreatment CGM features, comprising CGM features of the pretreatment CGM data, and a plurality of treatment CGM features, comprising CGM features of the treatment CGM data; training the machine learning model for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus, by feeding the model an input comprising the patient data and the pretreatment CGM data, and an output comprising the treatment CGM features.

[0044] The machine learning model may for example be a regression model, a decision tree model, such as random forest and / or XGBoost, and / or a machine learning ensemble. In other examples, the machine learning model may be a supervised learning model. The presently disclosed method may be a computer-implemented method.

[0045] The use of CGM data in the present disclosure provides a more accurate and personalized assessment of a patient's condition compared to that of the prior art. By continuously monitoring glucose levels, the system can capture real-time data on the patient's glycemic fluctuations, which can enable understanding the immediate and long-term impacts of different treatment options.

[0046] The use of CGM data can allow for the extraction of detailed features such as meal impact, time in target glycemic ranges, and hypoglycemic and hyperglycemic events. These features provide a comprehensive view of the patient's glucose control, which static blood glucose measurements or HbA1c values cannot fully capture. This realtime data can lead to more accurate predictions of treatment outcomes, enabling the system to recommend treatment options that are finely tuned to the patient's unique glycemic patterns.

[0047] Moreover, the inclusion of CGM features in the scoring and ranking of treatment options ensures that the recommendations are not only based on general guidelines but also on the patient's actual physiological responses. This results in a more effective and safer treatment plan by avoiding options that could cause harmful glycemic excursions and by selecting those that maintain glucose levels within the desired range.

[0048] Compared to for example multi-criteria decision analysis models, the present system's use of CGM data provides a significant improvement in the accuracy of treatment recommendations. This leads to better management of type 2 diabetes, as the system can dynamically adjust to changes in the patient's condition and provide updated recommendations as new CGM data is collected.

[0049] Description of Drawings

[0050] Fig. 1 shows a schematic flow chart outlining how the system for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus is arranged to transfer data between modules, according to an embodiment of the present disclosure.

[0051] Fig. 2 shows an example of CGM data and different CGM features, according to an embodiment of the present disclosure.

[0052] Fig. 3 shows an example of training data, for training of a machine learning model, according to an embodiment of the present disclosure.

[0053] Fig. 4 shows regression of a trained machine learning model for three different CGM features for a treatment option, according to an embodiment of the present disclosure.

[0054] Fig. 5 shows an example of density distribution of features with statistically significant differences between the two groups of data, according to an embodiment of the present disclosure.

[0055] Fig. 6 shows a flow chart outlining a process for arriving at a recommendation of a selected treatment option, according to an embodiment of the present disclosure.

[0056] Fig. 7 shows a flow chart of a computer-implemented method for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus, according to an embodiment of the present disclosure.

[0057] Detailed description

[0058] In a first aspect, the present disclosure relates to a system for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus.

[0059] The system typically comprises one or more of a processing unit; a memory; an interface; an input device. The system may for example be a portable system and / or a handheld system such as a mobile phone. In other examples, the system is a computing device.

[0060] It is a preference that the system comprises an input device that is arranged to comprise CGM data. Further, the input device is preferably arranged to obtain, such as measure, continuous glucose monitoring (CGM) data of the patient. In specific example, the input device may be a CGM device, that is arranged to measure, preferably continuously, blood glucose values.

[0061] The system preferably comprises a memory that comprises instructions that, when executed by the processing unit, carries out a method for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus. Preferably said method is as disclosed elsewhere herein.

[0062] In particular, the system is arranged to carry out a method that comprises a step of performing an initial assessment of the treatment options. Said step is typically based on patient data (also known as patient information). The patient data may have for example been obtained by the interface and / or patient data input means.

[0063] The patient data may for example comprise patient demographics, clinical data, previous or present health conditions, year of diabetes onset, and / or a target value for HbA1c.

[0064] The patient data may for example comprise patient demographics, clinical data, previous or present health conditions, year of diabetes onset, and / or a target value for HbA1c. The clinical data may for example comprise one or more measurements of markers for metabolic control (such as HbA1c, Fasting glucose, and / or C-peptide), kidney function (such as creatinine, glomerular filtration rate, and / or albuminuria) and liver function (such as ASAT, ALAT, PK, Albumin, and / or bilirubin).

[0065] Preferably, the patient data comprises information of whether the patient suffers from one or more of the following conditions: cardiovascular disease, heart failure, gastroparesis, intestinal disease, liver disease, osteopeniaAporosis, insulin resistance, critical ischemia, arterial foot ulcer, pitting oedema, frequent urinary tract infection, residual urine, genital Candida, multiple illnesses, and / or bladder cancer.

[0066] Alternatively, or additionally, the patient data may comprise information of whether the patient has suffered from an episode of acute pancreatitis, an episode of acute kidney failure, ketoacidosis, lactic acidosis and / or bladder cancer. Yet further, the patient data may alternatively or additionally comprise information of whether the patient has a known or a planned pregnancy, a need of special consideration to risk of hypoglycemia, such as a workplace where hypoglycemia would be hazardous, and / or a need of special consideration to patient cost, such as financial difficulties

[0067] The initial assessment may for example comprise a step of selecting treatment options, comprising excluding disadvantageous treatment options, such as treatment options that are expected to have side effects to the patients, such as treatment options that have contraindications.

[0068] The initial assessment may alternatively or additionally comprise a step providing an initial scoring to treatment options, such as the selected treatment options, wherein the initial scoring is based on expected effects of the (selected) treatment options for the patient, for example wherein the expected effects are based on treatment guidelines, for example personalized treatment guidelines, such as for treatment of Type 2 diabetes.

[0069] Thus, depending on the arrangement of the system, the initial assessment step may exclude a number of treatment options that are expected to have an adverse effect on the patient. Further, the initial assessment may weight, rank and / or score the selected treatment option, i.e. the non-excluded treatment options, for example based on guidelines for treatment of type 2 diabetes mellitus. Preferably, the scoring of the selected treatment options are based on the patient data, so that the scoring is a personalized scoring of the different selected treatment options.

[0070] It is a preference that the system comprises an input device that is arranged to comprise and / or obtain CGM data. Further, the input device is preferably arranged to obtain, such as measure, continuous glucose monitoring (CGM) data of the patient. In specific example, the input device may be a CGM device, that is arranged to measure, preferably continuously, blood glucose values. The input device may for example comprise CGM data that has been obtained from the patient, such as measured, during a number of recent days, for example seven recent days, such as the last seven days. In specific examples, the CGM device may comprise a sensing device, such as a sensing patch, and a data storage device, such as a mobile phone that is arranged to be in communication with the sensing device. Thus, the data storage device may be further arranged to communicate with the memory of the present system, thereby providing the measured CGM data for further processing.

[0071] Preferably the system is arranged to extract a plurality of CGM features from the CGM data. Thus the system may be arranged to read and / or analyze the CGM data in order to extract characteristics of glucose values that may contain information about the type 2 diabetes mellitus of the patient, such as the disease progression and / or severity of the disease.

[0072] The features of the CGM data, i.e. the CGM features, may for example comprise meal features, time in one or more glycemic ranges, hypoglycemic features, mean glucose features, hyperglycemic features, and / or fluctuation features. In one example, the features are as listed in one or more of Tables 1-6. _

[0073] In specific examples the system is arranged such that the meal features comprises one or more of meal delta, meal impact, meal peak value, pre meal value, post meal value, pre peak gradient, post peak gradient, meal area under curve, and / or meal time to peak, or an averaged value thereof.

[0074] In specific examples the system is arranged such that the time in glycemic ranges are calculated as a proportion of the CGM data that is within said one or more glycemic ranges respectively.

[0075] In specific examples the system is arranged such that the glycemic ranges comprises one or more hypoglycemic ranges, such as a glucose value lower than 3.9 mmol / l, a glucose value between 3.0 and 3.9 mmol / l, and / or a glucose value below 3.0 mmol / l.

[0076] In specific examples the system is arranged such that the glycemic ranges comprises one or more hyperglycemic ranges, such as a glucose value higher than 10.0 mmol / l, a glucose value between 10.0 and 13.9 mmol / l, and / or a glucose value above 13.9 mmol / l.

[0077] In specific examples the system is arranged such that the hypoglycemic features comprise information of occurrences of the CGM data that is below a hypoglycemic threshold value, such as the number of occurrences, the respective time, and / or the total time. Alternatively or additionally, hypoglycemic features may comprise root causes of the events (e.g. a hypoglycemic event), and / or the LBGI value.

[0078] In specific examples the system is arranged such that the hypoglycemic threshold value is 3.9 mmol / l and the severe hypoglycemic threshold is 3.0 mmol / l.

[0079] In specific examples the system is arranged such that the hyperglycemic features comprise information of occurrences of the CGM data that is above a hyperglycemic threshold value, such as the number of occurrences, the respective time, and / or the total time. Alternatively or additionally hyperglycemia features may comprise information of the root causes of the events (e.g. a hyperglycemic event), and / or the HBGI value.

[0080] In specific examples the system is arranged such that the hyperglycemic threshold is 10.0 mmol / l and the severe hyperglycemic threshold is 13.9 mmol / l.

[0081] In specific examples the system is arranged such that the mean glucose features comprise one or more of mean glucose value, glucose management indicator (GMI), eHbAlc, fasting glucose, and / or nightly glucose trend.

[0082] Preferably, the system is arranged to predict, for each (selected) treatment option, by a machine learning model that is trained to predict outcome CGM features based on the CGM feature, the outcome of each selected treatment option. The outcome CGM features may thereafter be used for deriving a CGM score for each treatment option.

[0083] The machine learning model(s) may for example be a supervised learning model. In specific examples, the machine learning model(s) is a regression model and / or a decision tree model, such as random forest and / or XG Boost.

[0084] Preferably, the system is arranged to carry out a step of scoring the selected treatment options, including a CGM scoring, and a price scoring. Advantageously, the CGM scoring and the price scoring is carried out in addition to an initial scoring, of each treatment option and / or selected treatment option. The system may for example be arranged such that the CGM scoring of each selected treatment option is based on the predicted outcome of the selected treatment option. The CGM scoring may for example be based on a hypo score, reflecting a risk associated with low glucose value of the patient, a hyper score, reflecting a risk of a high glucose value of the patient, and / or a fluctuation score, reflecting a maintenance burden of the patient.

[0085] The hypo score may for example be derived from low blood glucose index (LBGI), of the selected treatment option. The hyper score may for example be derived from high blood glucose index (HBGI), of the selected treatment option. The fluctuation score may be derived from glucose variability percentage (GVP) , of the selected treatment option.

[0086] Alternatively, the hypo score may for example be the low blood glucose index (LBGI) , of the selected treatment option. The hyper score may for example be the high blood glucose index (HBGI) , of the selected treatment option. The fluctuation score may be the glucose variability percentage (GVP), of the selected treatment option.

[0087] The hypo score, the hyper score and / or the fluctuation score may have been normalized, such as by an expert panel.

[0088] In specific examples of the present disclosure, the CGM scoring of each selected treatment option is the lowest score, such as highest risk, of the hypo score, the hyper score, and the fluctuation score. In this way, it is possible to avoid treatment options that have for example a single bad value that is overshadowed by two good values.

[0089] The system may be arranged such that each selected treatment option is predicted by a separate trained machine learning model. For example, wherein the outcome of each selected treatment option, of one or more CGM features, is predicted by a trained machine learning model, such as a different trained machine learning model.

[0090] The system may be arranged such to comprise a step of ranking the selected treatment options based on the initial scoring, the CGM scoring, and the price scoring. In specific examples, the system is arranged such that the step of prioritizing comprises selecting / recommending the selected treatment option with the highest initial scoring.

[0091] The system may for example be arranged such that the step of prioritizing may comprise ranking the selected treatment options from the selected treatment option with the highest initial scoring to the selected treatment option with the lowest initial scoring.

[0092] Further, in case multiple treatment options have the same, a similar and / or an equivalent, initial scoring, the ranking of said treatment options with the same initial scoring is from the highest CGM scoring to the lowest CGM scoring.

[0093] Yet further, in case multiple treatment options have the same, a similar and / or an equivalent, initial scoring and the same, a similar and / or an equivalent, CGM scoring, the ranking of said treatment options with the same, a similar and / or an equivalent, initial scoring and same CGM scoring is from the highest price scoring, such as a low price, to the lowest price scoring, such as a high price.

[0094] Thus, in specific examples, the ranking is arranged such that the initial scoring is prioritized over that of the CGM scoring, which is prioritized over that of the price scoring.

[0095] The system according to any one of the preceding claims, wherein the prioritizing comprises selecting / recommending the selected treatment option with the highest overall scoring, and wherein the initial scoring, the CGM scoring and the price scoring have been provided weights based on their respective importance.

[0096] The system may in specific embodiments be arranged to comprise a step of providing, by the interface, at least one recommended treatment option, wherein the recommended treatment options exclude combination of selected treatment options that are expected to have harmful drug interaction.

[0097] In specific examples, the system is arranged to comprise a step of providing at least one recommended treatment option comprising prioritizing the initial scoring, over the CGM scoring and / or the price scoring, of the selected treatment option. In specific examples, the at least one recommended treatment option comprises the highest ranked selected treatment option.

[0098] In specific examples, the at least one recommended treatment option comprises the highest ranked selected treatment option, or, wherein the selected treatment option does not obtain sufficient effect in the patient, such as a decreased HbA1c value below a target value, the highest ranked selected treatment option together with the highest ranked remaining selected treatment option that does not give rise to a harmful drug interaction.

[0099] In a second aspect, the present disclosure relates to a computer-implemented method for treatment decision support of a patient for a plurality of treatment options for treatment of diabetes, in particular for treatment of type 2 diabetes mellitus.

[0100] The method may comprise providing a plurality of treatment options for treatment of type 2 diabetes mellitus.

[0101] The method typically comprises performing an initial assessment of the treatment options, based on patient data. The initial assessment may for example be performed according to an initial assessment module.

[0102] In specific examples, the initial assessment may comprise selecting treatment options, comprising excluding disadvantageous treatment options, such as treatment options that are contradicted by one or more parameters in the patient data and / or expected to have contraindications for the patient.Thus, in one example, the step of excluding comprises excluding treatment options that is expected to result in a contraindication for the patient.

[0103] Alternatively or additionally, excluding disadvantageous treatment options may comprise excluding treatment options that are contradicted by one or more parameters in the patient data, such as previous / current health conditions or drug concurrent treatment. In specific embodiments of the present disclosure, the initial assessment and / or scoring may comprise eliminating, selecting and / or weighing the treatment options based on treatment guidelines. The treatment guidelines may be patient specific guidelines, i.e. the treatment guidelines may take into account one or more variables of the patient data. The treatment guidelines may for example comprise or consist of guidelines according to American Diabetes Association (ADA), American Association of Clinical Endocrinology (AACE), European Association for the Study of Diabetes (EASD), Advanced Technology and Treatments for Diabetes (ATTD), and / or World Health Organisation (WHO).

[0104] The method may comprise a step of providing an initial scoring to the selected treatment options, wherein the initial scoring is based on expected effects of the selected treatment options for the patient, such as based on treatment guidelines. The initial scoring may for example be based on patient specific guidelines for treatment of type 2 diabetes mellitus.

[0105] Thus, the step of providing an initial scoring may comprise providing, for each nonexcluded treatment option, a score based on patient specific guidelines for treatment of type 2 diabetes mellitus.

[0106] In one embodiment of the present disclosure, the method comprises a step of obtaining continuous glucose monitoring (CGM) data of the patient, comprising glucose values of the patient obtained during a predetermined time period.

[0107] The CGM data may for example have been obtained for, at least, 7 days. Preferably the CGM data has been obtained (such as measured) recently. The CGM data could for example have been obtained continuously for a period of time, such as a number of recent days, for example the last seven days.

[0108] Preferably, the method comprises a step of extracting a plurality of CGM features from the CGM data. The CGM features may for example comprise meal features, time in one or more glycemic ranges, hypoglycemic features, mean glucose features, hyperglycemic features, and / or fluctuation features. Meal features may for example comprise one or more of meal delta, meal impact, meal peak value, pre meal value, post meal value, pre peak gradient, post peak gradient, meal area under curve, and / or meal time to peak, or an averaged value thereof.

[0109] The time in glycemic ranges may for example be calculated as a proportion of the CGM data that is within said one or more glycemic ranges respectively.

[0110] The method may further comprise a step of predicting. Preferably this step comprises predicting, for each treatment option, by a trained machine learning model, the outcome of each selected treatment option.

[0111] The method may comprise a step of scoring the selected treatment options, including a CGM scoring, and a price scoring.

[0112] Further, the method may comprise a step of ranking the selected treatment options based on the initial scoring, the CGM scoring, and the price scoring;

[0113] Further, the method may comprise a step of providing at least one recommended treatment option, wherein the at least one recommended treatment option excludes combinations of selected treatment options that are expected to have harmful drug interaction.

[0114] Fig. 7 shows an outline of a flow chart of a computer-implemented method for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus, according to an embodiment of the present disclosure.

[0115] In particular, Fig. 7A shows a computer-implemented method for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus, according to an embodiment of the present disclosure. The method of this example is shown to comprise a step of performing an initial assessment of the treatment options (71).

[0116] The step of performing an initial assessment of the treatment options (71) is shown to comprise an initial assessment of the treatment options, based on patient data, wherein the initial assessment comprises: i. selecting treatment options, comprising excluding disadvantageous treatment options, such as treatment options that are contradicted by one or more parameters in the patient data; and ii. providing an initial scoring to the selected treatment options, wherein the initial scoring is based on expected effects of the selected treatment options for the patient, such as based on treatment guidelines.

[0117] Further, the method of the example shown in Fig. 7A comprises a step of obtaining CGM data (72), said step comprising obtaining continuous glucose monitoring (CGM) data of the patient, comprising glucose values of the patient obtained during a predetermined time period.

[0118] The method of the present example further comprises a step of extracting a plurality of CGM features from the CGM data (73), a step of predicting, for each treatment option, by a trained machine learning model, the outcome of each selected treatment option (74), a step of scoring the selected treatment options, including a CGM scoring, and a price scoring (75), a step of ranking the selected treatment options based on the initial scoring, the CGM scoring, and the price scoring (76). Typically, the CGM scoring is based on the outcome CGM features.

[0119] The method of the present disclosure further comprises a step of providing at least one recommended treatment option (77), wherein the recommended treatment options exclude combination of selected treatment options that are expected to have harmful drug interaction.

[0120] The computer-implemented method allows for an efficient method to provide an accurate treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus. In particular, the method is arranged to take into account current guidelines of diabetes treatment in combination with individual predicted outcomes. By the presently disclosed process, the optimal treatment is identified and presented to the user, typically the primary care physician, in a transparent way.

[0121] The method typically relies on an initial assessment module, a CGM module and a prioritization module, that, when combined, results in the efficient and accurate recommendation of a treatment option for the patient. Typically, the initial assessment module comprises excluding, selecting, and / or providing an initial scoring to, a number of treatment options, based on patient data. The method may for example comprise excluding treatment options that is expected to result in a contraindication for the patient. The method may comprise a step of providing an initial scoring comprising providing, for each non-excluded treatment option, a score based on patient specific guidelines for treatment of type 2 diabetes mellitus.

[0122] The patient data may for example comprise patient demographics, clinical data, previous or present health conditions, year of diabetes onset, and / or a target value for HbA1c. The clinical data may for example comprise one or more measurements of markers for metabolic control (such as HbA1c, Fasting glucose, and / or C-peptide), kidney function (such as creatinine, glomerular filtration rate, and / or albuminuria) and liver function (such as ASAT, ALAT, PK, Albumin, and / or bilirubin).

[0123] In the present example, the initial assessment module provides an output to the CGM module which is arranged to apply points to the selected treatment options, based on CGM features and the patient data. The CGM features may have been identified from CGM data of the patient that has been obtained continuously for a period of time, such as a number of recent days, for example the last seven days.

[0124] The CGM features are typically features that are indicative of the status of the type 2 diabetes mellitus of the patient, but may for example comprise the meal features, time in one or more glycemic ranges, hypoglycemic features, mean glucose features, hyperglycemic features, and / or fluctuation features.

[0125] Fig. 2 shows an example of a meal curve (CGM curve), from which meal features can be derived. In the figure the glucose level (mmol / l) is given as a function of the time since the start of the meal (min).

[0126] The meal features (which is a type of CGM feature) may include a meal pre value (8) which is the glucose value at the start of a meal, a meal post value (10) which is the value 2 h after start of a meal, a meal peak value (9) which is the maximum value during the meal interval. Impact (13) is the difference between the pre and peak values, and delta (15) is the difference between pre and post glucose values. Pre peak gradient (11) is the mean gradient between the pre and peak glucose values and post peak gradient (12) is the same between the peak and post glucose values. Meal AUC (16) is the area under the meal curve, and time to peak is the time difference between pre and peak times. Further, the mean of these features for all meals may typically be calculated for each individual.

[0127] The method typically comprises a step of predicting, for each treatment option, by a trained machine learning model, the outcome of each selected treatment option. Thus, the method may rely on multiple machine learning models, and wherein the result of a selected treatment option is predicted for each treatment option by a different machine learning model.

[0128] Further, the method typically comprises a step of scoring, comprising, for each selected treatment option, providing an initial score, a CGM score and / or a price score, wherein a better initial score means a better outcome for the patient according to the treatment guidelines, CGM scoring means a better outcome for the patient according to the machine learning model(s), and similarly a better pricing score means a lower price.

[0129] Based on the scoring of the selected treatment options, the method is arranged to provide a recommendation of the best selected treatment options that is not expected to have any harmful drug interactions.

[0130] In a further aspect, the present disclosure relates to a computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the steps of the computer-implemented method for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus as disclosed elsewhere herein.

[0131] In yet a further aspect, the present disclosure relates to a system for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus, wherein the system is arranged to carry out the computer-implemented method for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus as disclosed elsewhere herein. In even yet a further aspect, the present disclosure relates to a method of training a machine learning model for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus, the method comprising:

[0132] • obtaining patient data of a plurality of patients and CGM data of the patients before starting a new treatment and after starting said treatment, the CGM data comprising: i. pretreatment CGM data, comprising CGM data before the start of the treatment; ii. treatment CGM data, comprising CGM data after the start of the treatment, such as when an effective dose has been reached in the patient;

[0133] • extracting, for each patient, a plurality of pretreatment CGM features, comprising CGM features of the pretreatment CGM data, and a plurality of treatment CGM features, comprising CGM features of the treatment CGM data; training the machine learning model for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus, by feeding the model an input comprising the patient data and the pretreatment CGM data, and an output comprising the treatment CGM features.

[0134] The machine learning model may for example be a regression model and / or a decision tree model, such as random forest and / or XG Boost. In other examples, the machine learning model may be a supervised learning model.

[0135] The presently disclosed method may be a computer-implemented method.

[0136] The machine learning model may be trained to carry out the method for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus, as disclosed elsewhere herein. Preferably at least the step of predicting, for each treatment option, by a machine learning model, the outcome of each selected treatment option, such as the outcome CGM features, based on the CGM features.

[0137] Examples according to specific embodiments of the present disclosure

[0138] Example 1 : Training a model Materials and methods

[0139] Data that is obtained:

[0140] • 50+ patients

[0141] • 14-days CGM data prior to start of new treatment (pretreatment CGM data).

[0142] • 14-days CGM data after effective dose is reached of new treatment (treatment CGM data).

[0143] • Patient data (for example birth year, gender, pharmaceuticals, year of diabetes onset etc.).

[0144] A regression based decision tree-type model is trained. The model is evaluated using mean squared error, or a similar error method, on validation data.

[0145] The CGM data is divided into training data (80 %), validation data (10%), and test data (10%). The training and validation data will be determined using k-fold cross validation.

[0146] The CGM data comprise synthesized data for 100 individuals with type 2 diabetes for the three features fasting glucose, meal impact and meal delta, which are shown in Fig. 3A-C.

[0147] Specifically, Fig. 3A discloses the distribution of CGM data before start of the treatment (open rings) and after start of the treatment (dots) for the CGM feature fasting glucose.

[0148] Fig. 3B discloses the distribution of CGM data before start of the treatment (open rings) and after start of the treatment (dots) for the CGM feature meal impact.

[0149] Fig. 3C discloses the distribution of CGM data before start of the treatment (open rings) and after start of the treatment (dots) for the CGM feature meal delta.

[0150] CGM features are extracted from the CGM data, both pre and post start of new treatment. The pre features will be represented as a vector, X, and the post features as a vector Y.

[0151] Meal features

[0152] Time of meals in the CGM data is determined using a model for meal detection and the delta, impact, pre, post, peak and time to peak features are illustrated in Fig. 2. As shown in Fig. 2, the meal features (also referred to as CGM features) may include a meal pre value (8) which is the glucose value at the start of a meal, meal post value (10) is the value 2 h after start of a meal, meal peak value (9) is the maximum value during the meal interval. Impact (13) is the difference between the pre and peak values, and delta (15) is the difference between pre and post glucose values. Pre peak gradient (11) is the mean gradient between the pre and peak glucose values and post peak gradient (12) is the same between the peak and post glucose values. Meal AUG (16) is the area under the meal curve, and time to peak is the time difference between pre and peak times. The mean of these features for all meals is calculated for each individual.

[0153] The different glycemic range features are calculated as the proportion of time where the glucose value is between the specified intervals. These features are measured in percentages.

[0154] A hypoglycemic event is defined as when the glucose value is lower than 3.9 mmol / l. A mid hypoglycemic event is when the glucose value is between 3.9 and 3.0 mmol / l and a severe event is below 3.0 mmol / l. A hyperglycemic event is when the glucose level is above 10 mmol, mild is between 10 and 13.9 mmol / l and severe is above 13.9 mmol / l. The duration of the event is how long the glucose value(s) stays in the respective interval.

[0155] To determine the root cause of a hypo- or hyperglycemic event, a model will be used. The root causes for hyperglycemia can for example be lack of meal insulin or lack of basal insulin. LBGI and HBGI are features described in Kovatchev et al., 2019, "Glycemic Variability: Risk Factors, Assessment, and Control,", which are weighted with the help of a cost function to represent the non-linear risk associated with low values compared to high values.

[0156] The mean glucose value, standard deviation and CV are calculated according to the general statistical methods. The estimated HbA1c, eHbAlc, is calculated using the following formula, where gtis the glucose values in mmol / l: 6.9365 - 6.6282 Glucose management indicator, GMI, is another version of an estimation metric of HbA1c (Bergenstal et al., 2018, "Glucose Management Indicator (GMI): A New Term for Estimating A1C From Continuous Glucose Monitoring," ), which is calculated according to the following formula, where gtis the glucose values in mmol / l:

[0157] Glycemic variability percentage, GVP, is metric for variability which has been shown to better describe the fluctuation amplitude and frequency compared to standard deviation and CV (Peyser et al., 2018, "Glycemic Variability Percentage: A Novel Method for Assessing Glycemic Variability from Continuous Glucose Monitor Data,").

[0158] Fasting glucose values are determined as the mean of the glucose levels at 5 am, or prior to the first meal of the day if a meal is detected prior to 5 am. The nightly glucose trend is the change of glucose values during the night, when the individual is expected to be fasting and resting, which reflects the body’s endogenous insulin production and / or the long-acting insulin treatment.

[0159] The values of all these features from the pre-start-of-treatment CGM measurement, vector X, will be used as the input to the decision tree model, along with the patient data, vector P, such as year of birth, year of diabetes diagnosis, current treatment etc. The values of the features from the post-start-of-treatment, vector Y, will be used as labels, that the model should be trained to predict.

[0160] Results

[0161] Three separate decision tree models are trained, one for each feature, with a depth of 2 on the CGM data results. As shown in Fig. 4, decision tree regression of synthetic data for values prior to treatment and effect of treatment on the features fasting glucose (Fig. 4A), meal impact (Fig. 4B), and meal delta (Fig. 4C).

[0162] The models depend on patient data, such as duration of the diabetes, resulting in differences in the models for different patients with the same pretreatment values.

[0163] Conclusion The regression models can now easily be used to see the predicted effect of a certain treatment, and treatment options can thereby be compared to each other, and the most effective treatment can then be chosen for each patient.

[0164] Example 2: Use of machine learning models for predicting the outcome of selected treatment options

[0165] Materials and methods

[0166] Data that is received from the patient:

[0167] • Patient data (birth year, gender, pharmaceuticals, year of diabetes onset etc.).

[0168] • 14-days CGM data prior to start of new treatment.

[0169] One trained machine learning model per treatment option is used, wherein the machine learning model is trained according to Example 1.

[0170] The values of the CGM features from the CGM measurement, vector X, are used as input to the treatment option models, along with the patient data, vector P. The output of each model is a prediction of what the CGM feature values will be using an effective dose of that treatment option.

[0171] The scoring

[0172] To prioritize the predicted outcomes, in terms of CGM feature values, for the treatment options, a scoring system is used to score each outcome to reflect the benefit to the patient. The range of the scoring is 0-100, where a perfect glucose control and minimal maintenance burden for the patient is represented by the score of 100. The scoring system consists of three dimensions:Hypo score, Hyper score and Fluctuation score.

[0173] The hypo score is derived from the LBGI, low blood glucose index, and hyper score from HBGI, high blood glucose index, which reflects the risks associated with low and high glucose values as well as taking the increased risk of low values compared to high values into account. The fluctuation score is derived from GVP, glucose variability percentage, which better reflects the maintenance burden to the patient than other fluctuation metrics, for example standard deviation. These three values have been normalized using min-max normalization, where the min values are set to reflect the score of 100 and the max values are set to reflect the score of 0, according to a consensus decision of an endocrinology expert panel.

[0174] The overall score is set according to the lowest of these three scores, so as not to miss for example a bad hypo score due to it being overshadowed by good hyper and fluctuation scores.

[0175] Results

[0176] The scores of the predicted CGM features for each treatment option are easily sorted and the treatments are thereby prioritized. For the sake of transparency, the user of the system can view all three dimensions of the scoring, as well as all the predicted CGM feature values.

[0177] Conclusion

[0178] The treatment recommendation in addition to prior art be based on individual predicted outcomes of each treatment, thereby optimizing healthcare for the patient. for treatment decision of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus

[0179] Materials and methods

[0180] • Patient data (birth year, gender, pharmaceuticals, year of diabetes onset etc.).

[0181] • 14-days CGM data prior to start of new treatment is measured by a CGM device.

[0182] • One trained ML model per treatment option, wherein the ML models have been trained according to Example 1.

[0183] The patient data may for example comprise patient demographics, clinical data, previous or present health conditions, year of diabetes onset, and / or a target value for HbA1c. The clinical data may for example comprise one or more measurements of markers for metabolic control (such as HbA1c, Fasting glucose, and / or C-peptide), kidney function (such as creatinine, glomerular filtration rate, and / or albuminuria) and liver function (such as ASAT, ALAT, PK, Albumin, and / or bilirubin). The patient data also contains information about the current or previous treatments for diabetes and other current or previous treatments. Other life situations that are important to include are known or planned pregnancy, need of special consideration to risk of hypoglycemia, need of special consideration to patient cost.

[0184] The transfer of data to, from and / and within the system, according to an embodiment of the present disclosure, is described in Fig. 1. In particular, a plurality of treatment options (1) are provided together with the patient data (6) to a first initial assessment module (2).

[0185] In the initial assessment module, an initial assessment of the treatment options are made comprising: i. selecting treatment options, comprising excluding disadvantageous treatment options, such as treatment options that are contradicted by one or more parameters in the patient data, such as previous / current health conditions or drug concurrent treatment. Alternatively, or additionally, the initial assessment module may be arranged to provide an initial scoring to the selected treatment options, wherein the initial scoring is based on expected effects of the selected treatment options for the patient, such as based on treatment guidelines.

[0186] The selected treatment options are provided together with the patient data (6) to a CGM module (3). The CGM module is typically arranged to obtain, from the input device (e.g. a CGM device) CGM data. The CGM module is typically arranged to predict by a machine learning model the outcome of each selected treatment option. Typically, the outcome is the outcome CGM features. The machine learning model may for example be trained to predict the outcome CGM features based on the CGM features. The outcome CGM features may thereafter for example be used to obtain a CGM score, typically for each selected treatment option.

[0187] The CGM data may for example be used according to Example 3 in order to predict, by the CGM module, the outcome of the various treatment options, and to provide them with a CGM score.

[0188] Further, each treatment option is typically provided with a price score based on their price. This may be provided for example as part of the step of providing treatment options (1). The CGM scoring may for example be provided as part of the CGM module (3).

[0189] All selected treatment options are then typically prioritized and / or ranked (4). Typically in order of the initial assessment score, secondary according to the CGM score, and lastly according to the price score. However, the prioritization / ranking may be carried out in other orders as well.

[0190] In addition, the method comprise a step of providing (5), by the interface, at least one recommended treatment option, wherein the recommended treatment options exclude combination of selected treatment options that are expected to have harmful drug interaction.

[0191] This step may for example comprise a rule-based algorithm that is used to compile a recommendation, consisting of one or more treatment options, which also takes drug interactions into account, as can be seen in Fig. 6A-C.

[0192] For the recommendation, the first step is to order the option according to the initial assessment score, where treatment options A, C and D have an equal score, and B has a lower score. Next, the treatment options A, C and D are ranked according to the CGM-score, where A and C are highest but still equal. Lastly, A and C are ranked according to the price score. This results in a prioritization order of: C, A, D, B.

[0193] The patient is already taking treatment option C at max dosage, but the HbA1c value is higher than the target value, the next treatment option in the priority list (A) is added, as there is no harmful drug interaction between the two options.

[0194] The recommendation resulting from the system is presented to the user, along with comments about which prior-art-rules are activated, as well as the predicted CGM features for each treatment option. The user, which is mainly meant to be a primary care physician, is therefore able to see the reasoning behind the recommendations.

[0195] Conclusion

[0196] The system provides a transparent treatment recommendation based on measured CGM data of the patient and provides it to the user in a transparent way. Example 4: analysis of CGM feature differences for GLP-1 vs SGLT2

[0197] Introduction

[0198] This example is meant to illustrate the differences in effects on the CGM features between treatment options that the ML models are expected to reflect, supporting the belief that this system will serve to increase the benefit for the patient of an individualized treatment selection.

[0199] Materials and methods

[0200] CGM data from 25 patients who were prescribed either GLP-1 or SGLT2 as part of their treatment for diabetes mellitus type 2, was used.

[0201] The patient group was distributed as 17 individuals with GLP-1 as part of their treatment and 7 individuals with SGLT2. The GLP-1 group had a range of 18-52 days with available CGM data (31 ± 9 days) and the SGLT2 group had a range of 25-29 days with available CGM data (28 ± 1 days).

[0202] Features

[0203] From the CGM data, we extracted relevant CGM features, listed in Tables 1-6. Method In order to see which, if any, of the features were significantly different between the two groups of patients, we conducted a Mann-Whitney U test for the features with continuous values which suggests a significant difference (p-value < 0.05).

[0204] Results

[0205] In order to see which, if any, of the features were significantly different between the two groups of patients, we conducted a Mann-Whitney U test for the features with continuous values which suggests a significant difference (p-value < 0.05) in the following features:

[0206] • GVP

[0207] • CV

[0208] • Meal delta value

[0209] • Meal impact value

[0210] • Pre peak gradient

[0211] • Meal AUC The density distribution of these features for the two groups can be seen in Fig. 5. In particular, Fig. 5A shows the CGM feature GVP, Fig. 5B shows the CGM feature CV, Fig. 50 shows the CGM feature Meal delta value, Fig. 5D shows the CGM feature Meal impact value, Fig. 5E shows the CGM feature Pre peak gradient, Fig. 5F shows the CGM feature Meal AUC.

[0212] The results suggest that there is a difference of effect between the treatment options GLP-1 and SGLT2, supporting the claim that such differences can be predicted by ML models and used to select the treatment most beneficial to the patient.

Claims

Claims1. A system for treatment decision support of a patient for a plurality of treatment options for treatment of diabetes mellitus, such as type 2 diabetes mellitus, wherein the system comprises:• a processing unit;• a memory;• an interface;• an input device; wherein, the memory comprises instructions that, when executed by the processing unit, carries out a method comprising:I. performing an initial assessment of the treatment options, based on patient data, wherein the initial assessment comprises: i. selecting treatment options, comprising excluding disadvantageous treatment options, such as treatment options that are contradicted by one or more parameters in the patient data, such as previous / current health conditions or drug concurrent treatment; ii. providing an initial scoring to the selected treatment options, wherein the initial scoring is based on expected effects of the selected treatment options for the patient, such as according to current treatment guidelines;II. obtaining, by the input device, continuous glucose monitoring (CGM) data of the patient, comprising glucose values of the patient obtained during a predetermined time period;III. extracting a plurality of CGM features from the CGM data;IV. predicting, by a trained machine learning model, the outcome of each selected treatment option, such as the outcome CGM features;V. scoring the selected treatment options, including a CGM scoring, such as wherein the CGM scoring is based on the outcome CGM features, and a price scoring;VI. ranking the selected treatment options based on the initial scoring, the CGM scoring, and the price scoring;VII. providing, by the interface, at least one recommended treatment option based on the ranking of the selected treatment options, wherein the recommended treatment options excludes combination of selected treatment options that are expected to have harmful drug interaction;thereby providing the treatment decision support of the patient for the plurality of treatment options.

2. The system according to claim 1 , wherein the patient data comprises patient demographics, clinical data, previous or present health conditions, year of diabetes onset, and / or a target value for HbA1c.

3. The system according to claim 2, wherein the clinical data comprises one or more measurements of markers for metabolic control (such as HbA1c, Fasting glucose, and / or C-peptide), kidney function (such as creatinine, glomerular filtration rate, and / or albuminuria) and liver function (such as ASAT, ALAT, PK, Albumin, and / or bilirubin)4. The system according to any one of the preceding claims, wherein the patient data comprises information of whether the patient suffers from one or more of the following conditions: cardiovascular disease, heart failure, gastroparesis, intestinal disease, liver disease, osteopeniaAporosis, insulin resistance, critical ischemia, arterial foot ulcer, pitting oedema, frequent urinary tract infection, residual urine, genital Candida, multiple illnesses, and / or bladder cancer.

5. The system according to any one of the preceding claims, wherein the patient data comprises information of whether the patient has suffered from an episode of acute pancreatitis, an episode of acute kidney failure, ketoacidosis, lactic acidosis and / or bladder cancer.

6. The system according to any one of the preceding claims, wherein the patient data comprises information of whether the patient has a known or a planned pregnancy, a need of special consideration to risk of hypoglycemia, such as a workplace where hypoglycemia would be hazardous, and / or a need of special consideration to patient cost, such as financial difficulties.

7. The system according to any one of the preceding claims, wherein the step of excluding comprises excluding treatment options that are contradicted by one or more parameters in the patient data.

8. The system according to any one of the preceding claims, wherein the step of providing an initial scoring comprises providing, for each non-excluded treatment option, a score based on patient specific guidelines for treatment of diabetes mellitus, such as type 2 diabetes mellitus.

9. The system according to any one of the preceding claims, wherein the initial scoring comprises eliminating, selecting and / or weighing the treatment options based on treatment guidelines.

10. The system according to claim 9, wherein the treatment guidelines comprises indications and contraindications of the treatment options, based on the patient data.

11. The system according to any one of the preceding claims, wherein the input device is arranged to measure a blood glucose level of the patient.

12. The system according to any one of the preceding claims, wherein the input device is a CGM device.

13. The system according to any one of the preceding claims, wherein the step of obtaining CGM data comprises measuring CGM data, such as for a predetermined amount of time, such as at least seven days.

14. The system according to any one of the preceding claims, wherein the CGM data comprises glucose values of the patient that are obtained, such as measured, during a predetermined time period.

15. The system according to any one of the preceding claims, wherein the CGM data has been obtained continuously for, at least, seven days, such as the last 7 days.

16. The system according to any one of the preceding claims, wherein the CGM features comprises meal features, time in one or more glycemic ranges, hypoglycemic features, mean glucose features, hyperglycemic features, and / or fluctuation features.

17. The system according to claim 16, wherein the meal features comprises one or more of meal delta, meal impact, meal peak value, pre meal value, post meal value, pre peak gradient, post peak gradient, meal area under curve, and / or meal time to peak, or an averaged value thereof.

18. The system according to any one of claims 16-17, wherein the time in glycemic ranges are calculated as a proportion of the CGM data that is within said one or more glycemic ranges respectively.

19. The system according to any one of claims 16-18, wherein the glycemic ranges comprises one or more hypoglycemic ranges, such as a glucose value lower than 3.9 mmol / l, a glucose value between 3.0 and 3.9 mmol / l, and / or a glucose value below 3.0 mmol / l.

20. The system according to any one of claims 16-19, wherein the glycemic ranges comprises one or more hyperglycemic ranges, such as a glucose value higher than 10.0 mmol / l, a glucose value between 10.0 and 13.9 mmol / l, and / or a glucose value above 13.9 mmol / l.

21. The system according to any one of claims 16-20, wherein the hypoglycemic features comprise information of occurrences of the CGM data that is below a hypoglycemic threshold value, such as the number of occurrences, the respective time, and / or the total time, LBGI, and / or the root causes.

22. The system according to claim 21, wherein the hypoglycemic threshold value is 3.9 mmol / l.

23. The system according to any one of claims 16-22, wherein the hyperglycemic features comprise information of occurrences of the CGM data that is above a hyperglycemic threshold value, such as the number of occurrences, the respective time, and / or the total time, HBGI, and / or the root causes.

24. The system according to claim 23, wherein the hyperglycemic threshold is 10.0 mmol / l.

25. The system according to any one of claims 16-24, wherein the mean glucose features comprise one or more of mean glucose value, glucose management indicator (GMI), eHbAlc, fasting glucose, and / or nightly glucose trend.

26. The system according to any one of the preceding claims, wherein the CGM scoring, of each selected treatment option, is based on the predicted outcome of the selected treatment option, wherein the scoring is based on a hypo score, reflecting a risk associated with low glucose value of the patient, a hyper score, reflecting a risk of a high glucose value of the patient, and a fluctuation score, reflecting a maintenance burden of the patient.

27. The system according to claim 26, wherein the CGM scoring of each selected treatment option is the lowest score, such as highest risk, of the hypo score, the hyper score, and the fluctuation score.

28. The system according to any one of claims 26-27, wherein the hypo score, the hyper score and the fluctuation score have been normalized.

29. The system according to any one of claims 26-28, wherein the hypo score is derived from low blood glucose index (LBGI), the hyper score is derived from high blood glucose index (HBGI), and / or the fluctuation score is derived from glucose variability percentage (GVP).

30. The system according to any one of the preceding claims, wherein the outcome of each selected treatment option is predicted by a separate trained machine learning model.

31. The system according to any one of the preceding claims, wherein the outcome of each selected treatment option, of one or more CGM features, is predicted by atrained machine learning model, such as a different trained machine learning model.

32. The system according to any one of the preceding claims, wherein the machine learning model(s) is a supervised learning model.

33. The system according to any one of the preceding claims, wherein the machine learning model(s) is a decision tree model.

34. The system according to any one of the preceding claims, wherein the prioritizing comprises selecting the selected treatment option with the highest initial scoring.

35. The system according to any one of the preceding claims, wherein the prioritizing comprises ranking the selected treatment options from the selected treatment option with the highest initial scoring to the treatment option with the lowest initial scoring.

36. The system according to claim 35, wherein, in case multiple treatment options have the same, equivalent, or similar, initial scoring, the ranking of said treatment options with the same, a similar and / or an equivalent, initial scoring is from the highest CGM scoring to the lowest CGM scoring.

37. The system according to claim 36, wherein, in case multiple treatment options have the same, an equivalent, or a similar initial scoring and the same, an equivalent, or a similar CGM scoring, the ranking of said treatment options with the same, an equivalent, or a similar initial scoring and same CGM scoring is from the highest price scoring, such as a low price, to the lowest price scoring, such as a high price.

38. The system according to any one of the preceding claims, wherein the prioritizing comprises selecting the selected treatment option with the highest overall scoring, and wherein the initial scoring, the CGM scoring and the price scoring have been provided weights based on their respective importance.

39. The system according to any one of the preceding claims, wherein the step of providing at least one recommended treatment option comprises prioritizing the initial scoring, over the CGM scoring and / or the price scoring, of the selected treatment option.

40. The system according to any one of the preceding claims, wherein the recommended treatment options comprise, or consist of, the highest ranked selected treatment options.

41. The system according to any one of the preceding claims, wherein the at least one recommended treatment option comprise the highest ranked selected treatment options, or, wherein the selected treatment options does not obtain sufficient effectin the patient, such as a decreased HbA1c value below a target value, the highest ranked selected treatment option together with the highest ranked remaining selected treatment option that does not give rise to a harmful drug interaction.

42. A computer-implemented method for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus, the method comprising:• performing an initial assessment of the treatment options, based on patient data, wherein the initial assessment comprises: i. selecting treatment options, comprising excluding disadvantageous treatment options, such as treatment options that are contradicted by one or more parameters in the patient data; ii. providing an initial scoring to the selected treatment options, wherein the initial scoring is based on expected effects of the selected treatment options for the patient, such as based on treatment guidelines;• obtaining continuous glucose monitoring (CGM) data of the patient, comprising glucose values of the patient obtained during a predetermined time period;• extracting a plurality of CGM features from the CGM data;• predicting, for each treatment option, by a machine learning model, the outcome of each selected treatment option;• scoring the selected treatment options, including a CGM scoring, and a price scoring;• ranking the selected treatment options based on the initial scoring, the CGM scoring, and the price scoring;• providing at least one recommended treatment option based on the ranking of the selected treatment options, wherein the recommended treatment options exclude combination of selected treatment options that are expected to have harmful drug interaction; thereby providing the treatment decision support of the patient for the plurality of treatment options.

43. The method according to claim 42, wherein the patient data comprises patient demographics, clinical data, previous or present health conditions, year of diabetes onset, and / or a target value for HbA1c.

44. The method according to claim 43, wherein the clinical data comprises one or more measurements of markers for metabolic control (such as HbA1c, Fasting glucose, and / or C-peptide), kidney function (such as creatinine, glomerular filtration rate, and / or albuminuria) and liver function (such as ASAT, ALAT, PK, Albumin, and / or bilirubin).

45. The method according to any one of claims 42-44, wherein the patient data comprises information of whether the patient suffers from one or more of the following conditions: cardiovascular disease, heart failure, gastroparesis, intestinal disease, liver disease, osteopeniaAporosis, insulin resistance, critical ischemia, arterial foot ulcer, pitting oedema, frequent urinary tract infection, residual urine, genital Candida, multiple illnesses, and / or bladder cancer.

46. The method according to any one of claims 42-45, wherein the patient data comprises information of whether the patient has suffered from an episode of acute pancreatitis, an episode of acute kidney failure, ketoacidosis, lactic acidosis and / or bladder cancer.

47. The method according to any one of claims 42-46, wherein the patient data comprises information of whether the patient has a known or a planned pregnancy, a need of special consideration to risk of hypoglycemia, such as a workplace where hypoglycemia would be hazardous, and / or a need of special consideration to patient cost, such as financial difficulties.

48. The method according to any one of claims 42-47, wherein the step of excluding comprises excluding treatment options that is contradicted by one or more parameters in the patient data.

49. The method according to any one of claims 42-48, wherein the step of providing an initial scoring comprises providing, for each non-excluded treatment option, a score based on patient specific guidelines for treatment of type 2 diabetes mellitus.

50. The method according to any one of claims 42-49, wherein the initial scoring comprises eliminating, selecting and / or weighing the treatment options based on treatment guidelines.

51. The method according to claim 50, wherein the treatment guidelines comprises indications and contraindications of the treatment options, based on the patient data.

52. The method according to any one of claims 42-51 , wherein the CGM data has been obtained continuously for, at least, seven days, such as the last seven days.

53. The method according to any one of claims 42-52, wherein the features of the CGM data comprises meal features, time in one or more glycemic ranges, hypoglycemic features, mean glucose features, hyperglycemic features, and / or fluctuation features.

54. The method according to claim 53, wherein the meal features comprises one or more of meal delta, meal impact, meal peak value, pre meal value, post meal value, pre peak gradient, post peak gradient, meal area under curve, and / or meal time to peak, or an averaged value thereof.

55. The method according to any one of claims 53-54, wherein the time in glycemic ranges are calculated as a proportion of the CGM data that is within said one or more glycemic ranges respectively.

56. The method according to any one of claims 53-55, wherein the glycemic ranges comprises one or more hypoglycemic ranges, such as a glucose value lower than 3.9 mmol / l, a glucose value between 3.0 and 3.9 mmol / l, and / or a glucose value below 3.0 mmol / l.

57. The method according to any one of claims 53-56, wherein the glycemic ranges comprises one or more hyperglycemic ranges, such as a glucose value higher than 10.0 mmol / l, a glucose value between 10.0 and 13.9 mmol / l, and / or a glucose value above 13.9 mmol / l.

58. The method according to any one of claims 53-57, wherein the hypoglycemic features comprise information of occurrences of the CGM data that is below a hypoglycemic threshold value, such as the number of occurrences, the respective time, and / or the total time.

59. The method according to claim 58, wherein the hypoglycemic threshold value is 3.9 mmol / l.

60. The method according to any one of claims 53-59, wherein the hyperglycemic features comprise information of occurrences of the CGM data that is above a hyperglycemic threshold value, such as the number of occurrences, the respective time, and / or the total time.

61. The method according to claim 60, wherein the hyperglycemic threshold is 10.0 mmol / l.

62. The method according to any one of claims 53-61 , wherein the mean glucose features comprise one or more of mean glucose value, glucose management indicator (GMI), eHbAlc, fasting glucose, and / or nightly glucose trend.

63. The method according to any one of claims 42-62, wherein the CGM scoring, of each selected treatment option, is based on the predicted outcome of the selected treatment option, wherein the scoring is based on a hypo score, reflecting a risk associated with low glucose value of the patient, a hyper score, reflecting a risk of a high glucose value of the patient, and a fluctuation score, reflecting a maintenance burden of the patient.

64. The method according to claim 63, wherein the CGM scoring of each selected treatment option is the lowest score, such as highest risk, of the hypo score, the hyper score, and the fluctuation score.

65. The method according to any one of claims 63-64, wherein the hypo score, the hyper score and the fluctuation score have been normalized.

66. The method according to any one of claims 63-65, wherein the hypo score is derived from low blood glucose index (LBGI), the hyper score is derived from high blood glucose index (HBGI), and / or the fluctuation score is derived from glucose variability percentage (GVP).

67. The method according to any one of claims 42-66, wherein the outcome of each selected treatment option is predicted by a separate trained machine learning model.

68. The method according to any one of claims 42-67, wherein the outcome of each selected treatment option, is predicted by a trained machine learning model, such as a different trained machine learning model per selected treatment option.

69. The method according to any one of claims 42-68, wherein the machine learning model(s) is a supervised learning model.

70. The method according to any one of claims 42-69, wherein the machine learning model(s) is a decision tree model.

71. The method according to any one of claims 42-70, wherein the prioritizing comprises selecting the selected treatment option with the highest initial scoring.

72. The method according to any one of claims 42-71 , wherein the prioritizing comprises ranking the selected treatment options from the selected treatment option with the highest initial scoring to the selected treatment option with the lowest initial scoring.

73. The method according to claim 72, wherein, in case multiple treatment options have the same, equivalent, or similar initial scoring, the ranking of said treatment options with the same, a similar and / or an equivalent, initial scoring is from the highest CGM scoring to the lowest CGM scoring.

74. The method according to claim 73, wherein, in case multiple treatment options have the same, equivalent, or similar initial scoring and the same, a similar and / or an equivalent, CGM scoring, the ranking of said treatment options with the same, equivalent, or similar initial scoring and same, a similar and / or an equivalent, CGM scoring is from the highest price scoring, such as a low price, to the lowest price scoring, such as a high price.

75. The method according to any one of claims 42-74, wherein the prioritizing comprises selecting the selected treatment option with the highest overall scoring, and wherein the initial scoring, the CGM scoring and the price scoring have been provided weights based on their respective importance.

76. The method according to any one of claims 42-75, wherein the step of providing at least one recommended treatment option comprises prioritizing the initial scoring, over the CGM scoring and / or the price scoring, of the selected treatment option.

77. The method according to any one of claims 42-76, wherein the at least one recommended treatment option comprises the highest ranked selected treatment option.

78. The method according to any one of claims 42-77, wherein the at least one recommended treatment option comprises the highest ranked selected treatment option, or, wherein the selected treatment option does not obtain sufficient effect in the patient, such as a decreased HbA1c value below a target value, the highest ranked selected treatment option together with the highest ranked remaining selected treatment option that does not give rise to a harmful drug interaction.

79. A computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the steps of the method of any one of claims 42-78.

80. A system for treatment decision support of a patient for a plurality of treatment options for treatment of diabetes mellitus, such as type 2 diabetes mellitus, wherein the system comprises:• a processing unit;• a memory;• an interface;• an input device; wherein, the memory comprises instructions which, when the instructions are executed by the processing unit, carries out the steps of the method of any one of claims 42-78.

81. A method of training a machine learning model for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus, the method comprising:• obtaining patient data of a plurality of patients and CGM data of the patients before starting a new treatment and after starting said treatment, the CGM data comprising: i. pretreatment CGM data, comprising CGM data before the start of the treatment; ii. treatment CGM data, comprising CGM data after the start of the treatment;• extracting, for each patient, a plurality of pretreatment CGM features, comprising CGM features of the pretreatment CGM data, and a plurality of treatment CGM features, comprising CGM features of the treatment CGM data;• training the machine learning model for treatment decision support of a patient for a plurality of treatment options for treatment of type 2 diabetes mellitus, by feeding the model an input comprising the patient data and the pretreatment CGM data / features, and an output comprising the treatment CGM features.

82. The method according to claim 81 , wherein the plurality of patients comprises at least 50 patients.

83. The method according to any one of claims 81-82, wherein the machine learning model(s) is a supervised learning model.

84. The method according to any one of claims 81-83, wherein the machine learning model(s) is a regression model and / or a decision tree model, such as random forest and / or XGBoost.

85. The method according to any one of claims 81-84, wherein the pretreatment CGM data comprises CGM data during at least 7 days before start of the treatment.

86. The method according to any one of claims 81-85, wherein the treatment CGM data comprises CGM data during at least 7 days after start of the treatment, such as when an effective dose has been reached.

87. The method according to any one of claims 81-86, wherein the start of a new treatment is when an effective dose of the treatment has been reached in the patient.

88. The method according to any one of claims 81-87, wherein the patient data comprises patient demographics, clinical data, previous or present health conditions, year of diabetes onset, and / or a target value for HbA1c.

89. The method according to claim 88, wherein the clinical data comprises one or more measurements of markers for metabolic control (such as HbA1c, Fasting glucose, and / or C-peptide), kidney function (such as creatinine, glomerular filtration rate, and / or albuminuria) and liver function (such as ASAT, ALAT, PK, Albumin, and / or bilirubin).

90. The method according to any one of claims 81-89, wherein the patient data comprises information of whether the patient suffers from one or more of the following conditions: cardiovascular disease, heart failure, gastroparesis, intestinal disease, liver disease, osteopeniaAporosis, insulin resistance, critical ischemia, arterial foot ulcer, pitting oedema, frequent urinary tract infection, residual urine, genital Candida, multiple illnesses, and / or bladder cancer.

91. The method according to any one of claims 81-90, wherein the patient data comprises information of whether the patient has suffered from an episode of acute pancreatitis, an episode of acute kidney failure, ketoacidosis, lactic acidosis and / or bladder cancer.

92. The method according to any one of claims 81-91 , wherein the patient data comprises information of whether the patient has a known or a planned pregnancy, a need of special consideration to risk of hypoglycemia, such as a workplace where hypoglycemia would be hazardous, and / or a need of special consideration to patient cost, such as financial difficulties.

93. The method according to any one of claims 81-92, wherein the method is a computer-implemented method.

94. A machine learning model for treatment decision support, wherein the machine learning model has been trained according to any one of claims 81-93.

95. The machine learning model according to claim 94, wherein the model is a regression model and / or a decision tree model, such as random forest and / or XGBoost.

Images