Injection site defect detection

Abstract

Systems, methods, and devices are provided for predicting an infusion site condition. The techniques include applying a regression model to physiological glucose data and insulin delivery data to generate predictive data, running a trained machine learning model to process the predictive data and generate an output, and determining that the infusion site has failed or is likely to have failed based on the output.

Description

[Technical Field]

[0001] The present disclosure relates to detecting problems at a drug infusion site. [Background technology]

[0002] Subcutaneous insulin replacement therapy has proven to be the regimen of choice for controlling diabetes. Insulin is administered either via multiple daily injections or an infusion pump, with dosage informed by capillary glucose measurements performed several times daily with a blood glucose meter. This traditional approach is known to be imperfect because of the potential for large day-to-day (and indeed moment-to-moment) variations. Furthermore, this approach can be burdensome for patients, requiring repeated finger pricks, close monitoring of food intake, and careful control of insulin delivery. The advent of glucose measurement devices, such as continuous glucose monitors, creates the possibility of developing insulin delivery systems, such as closed-loop systems, that can automatically calculate and adjust insulin delivery in response to measured glucose levels. Summary of the Invention

[0003] Insulin delivery systems may deliver a calculated amount of insulin using an infusion set coupled to a patient's body at an infusion site. However, over time, the site where the infusion set is positioned may become less effective and malfunction. Accordingly, certain embodiments of the present disclosure are directed to systems, methods, and devices for detecting infusion site malfunction.

[0004] In Example 1, a method for predicting an infusion site condition includes applying a regression model to physiological glucose data and insulin delivery data to generate predictive data. The method further includes operating a trained machine learning model to process the predictive data and generate an output. The method further includes determining, based on the output, that the infusion site has failed or is likely to have failed.

[0005] In Example 2, the method of Example 1 further includes generating a warning signal indicating that the injection site has failed, and displaying a warning on a graphical user interface in response to the warning signal.

[0006] In Example 3, the method of Example 1 or 2, wherein the physiological glucose data and insulin delivery data are compiled over a period beginning when the infusion site is first used.

[0007] In Example 4, the method of Example 3, further comprising calculating a linear regression based on the aggregated physiological glucose data and insulin delivery data, wherein the predicted data is based at least in part on the linear regression.

[0008] In Example 5, the method of any one of Examples 1-4, wherein the output is a value indicative of a potential injection site malfunction.

[0009] In Example 6, the method of any one of Examples 1-5, wherein the physiological glucose data and insulin delivery data are updated and aggregated according to a set schedule.

[0010] In Example 7, the method of any one of Examples 1-6, wherein the prediction data includes p-values ​​for the selected metrics.

[0011] In Example 8, the method of Example 7, wherein the selected metric comprises a metric selected from a category of metrics including physiological glucose variability and physiological glucose.

[0012] In Example 9, the method of Example 7 or 8, wherein the selected metrics include time above range, low glucose index, total bolus, standard deviation, coefficient of variation, and instability index.

[0013] In Example 10, the method of any one of Examples 1 to 9, wherein the trained machine learning model is an XGBoost model.

[0014] In Example 11, the trained machine learning model is customized for the patient by retraining the machine learning model using the patient's previous physiological glucose data and insulin delivery data. The method of any one of Examples 1 to 10.

[0015] In Example 12, the method of any one of Examples 1 to 11, further comprising updating a status icon associated with the injection site on a user interface based on the output.

[0016] In Example 13, a computer program product comprising instructions for causing one or more processors to perform the steps of the method according to Examples 1-12.

[0017] In Example 14, a computer-readable medium storing the computer program product of Example 13.

[0018] In Example 15, a computer comprising the computer-readable medium of Example 14.

[0019] In Example 16, a method for predicting an infusion site condition includes applying, using an electronic controller, a regression model to physiological glucose data and insulin delivery data to generate predictive data. The method further includes using the electronic controller to operate a trained machine learning model to process the predictive data and generate an output. The method further includes using the electronic controller to determine, based on the output, that the infusion site has failed or is likely to have failed.

[0020] In Example 17, the method of Example 1 further includes generating a warning signal indicating that the injection site has failed, and displaying a warning on a graphical user interface in response to the warning signal.

[0021] In Example 18, the method of any one of Examples 1-17, wherein the physiological glucose data and insulin delivery data are compiled over a period beginning when the infusion site is first used.

[0022] In Example 19, the method of Example 3, further comprising calculating a linear regression based on the aggregated physiological glucose data and insulin delivery data, wherein the predicted data is based at least in part on the linear regression.

[0023] In Example 20, the non-transitory computer-readable medium includes instructions to cause a hardware processor to (1) apply a regression model to physiological glucose data and insulin delivery data to generate predictive data, (2) operate a trained machine learning model to process the predictive data and generate an output, and (3) determine, based on the output, that an infusion site has failed or is likely to have failed.

[0024] In Example 21, the output is a value indicative of a likelihood of infusion site failure.

[0025] In Example 22, the non-transitory computer-readable medium of Example 20, wherein the selected metric comprises a metric selected from a category of metrics including physiological glucose variability and physiological glucose.

[0026] In Example 23, the non-transitory computer-readable medium of Example 20, wherein the trained machine learning model is customized for the patient by retraining the machine learning model using the patient's previous physiological glucose data and insulin delivery data.

[0027] In Example 24, a system includes a controller including a processor and a memory. The memory stores instructions that cause the processor to apply a regression algorithm to the physiological glucose data and the insulin delivery data to generate predictive data, to operate a trained machine learning model to process the predictive data and generate an output, and to determine, based on the output, that an infusion site has failed or is likely to have failed.

[0028] In Example 25, the system of Example 24, wherein the output is a value indicative of a likelihood of injection site failure.

[0029] In Example 26, the instructions further cause the processor to generate a warning signal indicating that the infusion site has failed when the likelihood exceeds a threshold. The system further includes a user interface configured to receive the warning signal and, in response, generate a warning on the user interface.

[0030] In Example 27, the system of Example 24, wherein the prediction data includes p-values ​​for the selected metrics.

[0031] In Example 28, the system of Example 27, wherein the selected metric includes a metric selected from a category of metrics, the category including physiological glucose variability and physiological glucose.

[0032] In Example 29, the system of Example 27, wherein one of the selected metrics is average glucose.

[0033] In Example 30, the system of Example 24, wherein the physiological glucose data includes summarized data based on raw glucose data.

[0034] In Example 31, the system of Example 24, wherein the trained machine learning model is customized for the patient by retraining the machine learning model using the patient's previous physiological glucose data and previous insulin delivery data.

[0035] In Example 32, the system described in Example 24, wherein the controller further includes a rule-based algorithm.

[0036] In Example 33, the system of Example 24, wherein the rule-based algorithm is invoked after a predetermined period of time.

[0037] In Example 34, the system of Example 24 further includes a medication delivery device configured to deliver insulin to the patient and generate insulin delivery data.

[0038] In Example 35, the system of Example 34 further includes a glucose measuring device in communication with the controller and configured to generate physiological glucose data.

[0039] In Example 36, a non-transitory computer-readable medium comprising instructions that cause a hardware processor to perform a process described herein. [Brief explanation of the drawings]

[0040] [Figure 1] 1 shows a schematic diagram of a system for controlling physiological glucose, according to certain embodiments of the present disclosure. [Figure 2] 1 illustrates an infusion set according to certain embodiments of the present disclosure. [Figure 3] FIG. 1 shows a block diagram of a method for detecting infusion site malfunction using a model-based approach, according to certain embodiments of the present disclosure. [Figure 4] 1 shows a graph of raw and processed glucose data according to certain embodiments of the present disclosure. [Figure 5] FIG. 1 shows a block diagram of a method for detecting infusion site malfunction using a rule-based approach, according to certain embodiments of the present disclosure. [Figure 6]1 illustrates logic that may be used as part of a rules-based approach to detecting infusion site malfunction, according to certain embodiments of the present disclosure. [Figure 7] 1 illustrates logic that may be used as part of a rules-based approach to detecting infusion site malfunction, according to certain embodiments of the present disclosure. [Figure 8] 1 illustrates logic that may be used as part of a rules-based approach to detecting infusion site malfunctions, according to certain embodiments of the present disclosure. [Figure 9] 1 shows a graph illustrating area under the curve calculation in accordance with certain embodiments of the present disclosure. [Figure 10] FIG. 1 shows a block diagram of a method for detecting infusion site malfunctions using both model-based and rule-based approaches, according to certain embodiments of the present disclosure.

[0041] While the invention is susceptible to various modifications and alternative forms, specific embodiments have been shown by way of example in the drawings and are described in detail below. The intention, however, is not to limit the disclosure to the particular embodiments described, but instead to cover all modifications, equivalents, and alternatives falling within the scope of the appended claims. DETAILED DESCRIPTION OF THE INVENTION

[0042] Insulin delivery systems, such as closed-loop systems, automatically calculate and adjust insulin delivery doses in response to measured glucose levels. The systems deliver the calculated insulin delivery dose using an infusion set coupled to the patient's body. Over time, the site where the infusion set is positioned may become less effective, resulting in site failure. For example, failure may be due to inflammation or other tissue degradation at the site that reduces the effectiveness of insulin delivered to the site. Other examples of failure include site leakage, hyperglycemia and blood ketones, blood in the tubing, etc. If an infusion site fails and is not addressed (e.g., by replacing the infusion set with a new one and / or using a different infusion site), the patient may experience long-term hyperglycemia. Accordingly, certain embodiments of the present disclosure are directed to systems, methods, and devices for detecting infusion site failure.

[0043] The following description discloses various approaches to detecting infusion site malfunctions. These approaches include model-based approaches, rule-based approaches, and combinations of model-based and rule-based approaches. Before describing these approaches in detail, the description herein outlines examples of systems in which these approaches may be incorporated. The approaches described herein may also be utilized in other systems.

[0044] System Hardware 1 shows an exemplary representative block diagram of a system 10 for controlling physiological glucose. System 10 includes a drug delivery device 12, such as an infusion pump, that is removably coupled to a patient 14 via an infusion set 18. Drug delivery device 12 includes at least one drug reservoir 16 that contains a drug, such as insulin, although other suitable drugs may be delivered by system 10. Drug delivery device 12 may deliver the drug to patient 14 via infusion set 18, which provides a fluid pathway from drug delivery device 12 to patient 14. In other embodiments, delivery device 12 may include an infusion catheter coupled directly to the patient's subcutaneous tissue at an infusion site, without the use of an infusion set.

[0045] FIG. 2 shows an exemplary infusion set 18. The infusion set 18 includes a first proximal end 20 that communicates with the infusion pump's drug reservoir 16 (FIG. 1) to receive the drug and a second distal end 22 that communicates with the patient 14 to deliver the drug. At the first end 20, the infusion set 18 includes a reservoir connector 24 configured to mate with an insulin reservoir, flexible line set tubing 26, and a base connector 28 in the form of a male buckle portion. At the second end 22, the infusion set 18 includes an infusion base 30 in the form of a female buckle portion configured to receive the base connector 28, an adhesive pad 32 configured to adhere the infusion base 30 to the patient's skin, and an infusion catheter 34 (e.g., a needle or cannula) configured to be inserted into the patient's skin. In use, the drug is directed from the drug delivery device 12 through the line set tubing 26, through the infusion catheter 34, and into the patient's subcutaneous tissue. The infusion set 18 of FIG. 2 is merely one example of various types of infusion sets that may be used in the system 10.

[0046] Referring back to FIG. 1 , system 10 also includes an analyte sensor, such as a glucose measuring device 36. Glucose measuring device 36 may be a stand-alone device or an ambulatory device. One example of a glucose measuring device is a continuous glucose monitor (CGM). In a specific embodiment, glucose measuring device 36 may be a glucose sensor such as a Dexcom G6 Series Continuous Glucose Monitor, although any suitable continuous glucose monitor may be used. Glucose measuring device 36 is illustratively worn by patient 14 and includes one or more sensors that communicate with or monitor a physiological space (e.g., interstitial or subcutaneous space) within patient 14 and can sense an analyte (e.g., glucose) concentration in patient 14. In some embodiments, glucose measuring device 36 reports a value associated with the concentration of glucose in interstitial fluid, e.g., interstitial glucose. Glucose measuring device 36 may transmit a signal representing the interstitial glucose value to various other components of system 10.

[0047] System 10 includes a user interface device 38 (hereinafter "UI 38") that can be used to input user data into system 10, modify values, and receive information, prompts, data, etc. generated by system 10. In certain embodiments, UI 38 is a handheld user device specially programmed for system 10, or may be implemented via an application or app running on medication delivery device 12 or a personal smart device such as a phone, tablet, or watch. UI 38 may include input devices 40 (e.g., buttons, switches, icons) and a display 42 that displays a graphical user interface. A user may interact with input devices 40 and display 42 to provide information (e.g., alphanumeric data) to system 10. In certain embodiments, input devices 40 are icons (e.g., dynamic icons) on display 42 (e.g., a touchscreen). In one example, a patient uses UI 38 to announce events such as meals, the start of exercise, the end of exercise, and emergency stops.

[0048] The system 10 also includes an electronic controller 44. Although the controller 44 is shown as separate from the medication delivery device 12 and the UI 38, the controller 44 may be physically incorporated into either the medication delivery device 12 or the UI 38, or may be executed by a remote server. Alternatively, the UI 38 and the medication delivery device 12 may each include a controller 44, and control of the system 10 may be shared between the two controllers 44. For example, some functions and processes described herein may be performed by a controller that is part of a remote server, while other functions and processes are performed by a controller that is part of the UI 38. Regardless of its physical location within the system 10, the controller 44 is shown as being communicatively coupled, directly or indirectly, to the medication delivery device 12, the glucose measuring device 36, and the UI 38.

[0049] The controller 44 may include or be communicatively coupled to one or more interfaces 46 for communicatively coupling to the medication delivery device 12, the glucose measuring device 36, and / or the UI 38 via one or more communication links 48. Exemplary interfaces 46 include wired and wireless signal transmitters and receivers. Exemplary communication links 48 include wired communication links (e.g., serial communication), wireless communication links such as short-range wireless links such as Bluetooth, IEEE 802.11, proprietary wireless protocols, and / or the like. The term “communication link” may refer to the ability to communicate any type of information in at least one direction between at least two devices. The communication link 48 may be a persistent communication link, an intermittent communication link, an ad-hoc communication link, and / or the like. Information (e.g., pump data, glucose data, drug delivery data, user data) may be transmitted via the communication link 48. The medication delivery device 12, the glucose measuring device 36, and / or the UI 38 may also include one or more interfaces for communicatively coupling to other devices in the system 10, such as a remote server, via one or more communication links 48.

[0050] The controller 44 illustratively may include at least one processor 50 (e.g., a microprocessor) that executes software (e.g., software modules) and / or firmware stored in the memory 52 of the controller 44 and that is communicatively coupled to one or more interfaces 46 and to each other. The software / firmware code includes instructions that, when executed by the processor 50, cause the controller 44 to perform the process functions and functions described herein. Alternatively or additionally, the controller 44 may include one or more application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), hardwired logic, or combinations thereof. The memory 52 may include computer-readable storage media (e.g., non-transitory computer-readable media) in the form of volatile and / or non-volatile memory, which may be removable, non-removable, or a combination thereof. In embodiments, the memory 52 stores executable instructions 54 (e.g., computer code, machine-usable instructions, etc.) that cause the processor 50 to implement aspects of the system component embodiments discussed herein and / or execute aspects of the method and procedure embodiments discussed herein. The interface 46, the processor 50, and the memory 52 may be communicatively coupled by one or more buses. The memory 52 of the controller 44 is any suitable computer-readable medium accessible by the processor. The memory 52 may be a single storage device or multiple storage devices, may be located internal or external to the controller 44, and may include both volatile and non-volatile media.Exemplary memory includes random-access memory (RAM), read-only memory (ROM), electrically erasable programmable ROM (EEPROM), flash memory, CD-ROM, Digital Versatile Disk (DVD) or other optical disk storage, magnetic storage devices, or any other suitable medium configured to store data and accessible by controller 44.

[0051] In the depicted embodiment, the controller 44 receives information from multiple components of the system 10 and feeds that information (e.g., pump data, glucose data, drug delivery data, user data) into a control algorithm that determines at least one drug delivery control parameter that may, in part, govern the operation of the drug delivery device 12. In some specific embodiments, the controller 44 may receive pump data from the drug delivery device 12, glucose data from the glucose measuring device 36, and user data from the UI 38. The received pump data may include drug delivery data corresponding to a drug dose delivered to the patient 14 by the drug delivery device 12. The pump data may be provided by the drug delivery device 12 as the dose is delivered or on a predetermined schedule. The glucose data received by the controller 44 may include glucose concentration data from the glucose measuring device 36. The glucose data may be provided at a continuous rate, occasionally, or at predefined intervals (e.g., every 5 or 10 minutes).

[0052] The pump data, glucose data, drug delivery data, and user data may be provided to the controller 44 as they are acquired, provided on a predefined schedule, or queued in memory 52 and provided to the controller 44 upon request. User data may be entered into the UI 38 in response to user / patient prompts generated by the UI 38 and / or stated by the patient 14 as instructed during training. In some embodiments, at least a portion of the pump data, glucose data, and / or user data may be retrieved from memory 52 associated with the controller 44, and a portion of this data may be retrieved from memory within the medication delivery device 12.

[0053] At least one drug delivery parameter determined by the controller 44 may be a drug dose, which may at least partially govern drug administration to the patient 14 via the drug delivery device 12. In the case of insulin delivery (e.g., delivery of a fast-acting insulin or an ultrafast-acting insulin), the drug delivery parameter may be a basal rate (e.g., a basal profile including a predetermined time-varying insulin flow rate over a 24-hour period), a micro-bolus dose (e.g., a dose corrected for the basal rate), and / or a meal bolus. Basal delivery is the continuous delivery of insulin at a basal rate needed by the patient to maintain a desired level of glucose in the patient's blood outside of post-meal periods. A user may occasionally require a larger amount of insulin due to changes in activities such as eating a meal or other activities that affect the user's metabolism. This larger amount of insulin is referred to herein as a bolus. A meal bolus is generally a specific amount of insulin delivered over a short period of time. The nature of the drug delivery device 12 may require delivering a bolus as a continuous stream of insulin over a period of time or as a series of smaller, discrete insulin volumes delivered over a period of time. The meal bolus helps maintain glucose levels because the digestive system delivers large amounts of glucose to the bloodstream.

[0054] The term physiological glucose herein refers to the measured concentration of glucose in the body. In some embodiments, physiological glucose can be the concentration of glucose in blood, which may be referred to as blood glucose. In other embodiments, physiological glucose can be the concentration of glucose in plasma, which may be referred to as plasma glucose. Plasma glucose measurements are typically higher than blood glucose because blood is depleted of blood cells. The relationship between plasma glucose and blood glucose depends on hematocrit and can vary from patient to patient and over time.

[0055] Injection site defect detection As discussed above, over time, infusion sites can become less effective and fail. Failure can result in and be manifested as one or more of the following: leakage, hyperglycemia and ketones, blood in the ducts, inflammation at the infusion site, etc. However, removing an infusion set earlier than normal can increase the annual cost of the system. In addition to reducing annual costs, longer-lasting infusion sets improve the patient experience and overall convenience of using the system. Therefore, there is a desire and need for detecting infusion site failure.

[0056] This disclosure describes different approaches to detecting infusion site malfunctions. These approaches include model-based approaches, rule-based approaches, and a combination of model- and rule-based approaches. In certain embodiments, these approaches are used in near real time while the patient is using the infusion pump and / or infusion set being analyzed. In such embodiments, the controller 44 and / or other processing device may execute one or more of these approaches to alert the patient and / or their healthcare provider when the infusion site has malfunctioned or is likely to malfunction. The controller 44 may be part of the patient's smartphone and / or a remote server that operates functions requiring more computing resources. In other embodiments, these approaches may be used after removal of the infusion set to help identify the root cause of a hyperglycemic event. For example, instead of providing near real-time alerts, the approach may be used to generate a report classifying the root cause and severity level of hyperglycemic events and whether they are related to an infusion site malfunction or other causes, such as a missed or low meal bolus.

[0057] Although this disclosure refers to detecting infusion site malfunctions in systems that use infusion sets, the systems and methods herein may also be used to detect infusion site malfunctions when the infusion catheter of a delivery device is directly coupled to the patient's subcutaneous tissue without the use of an infusion set.

[0058] Model-based methods One approach to detecting infusion site malfunction is referred to herein as a model-based approach. Briefly, a model-based approach applies a trained machine learning model that outputs a prediction of whether a malfunction has occurred based on statistics derived from physiological glucose data and insulin delivery data. The model operates such that predictions regarding the infusion site are generated periodically. In certain embodiments, the model operates approximately once per hour, although other suitable periods or intervals may be implemented. While longer or shorter periods may be used, a period such as one hour may reduce or minimize the power consumed by running the model while obtaining a prediction for a period that may help prevent prolonged hyperglycemia. If a possible infusion site malfunction is predicted, a warning message may be sent to the patient and / or their healthcare provider.

[0059] FIG. 3 outlines an exemplary method 100 for detecting infusion site malfunction using a model-based approach. Method 100 may be executed by processor 50 of controller 44 based on, for example, pump data, glucose data, drug delivery data, and / or user data. In step 102, the patient's historical physiological glucose data and / or insulin delivery data are processed using a regression-based model or algorithm. As described above, the physiological glucose data may be generated by and received from a sensor (e.g., a CGM coupled to the patient), and the insulin delivery data may be generated by and received from a medical delivery device (e.g., a pump coupled to the patient). The physiological glucose data and / or insulin delivery data are aggregated and processed for each new time period beginning when the infusion set is first used and ending at the latest time for which the model is running. For example, if a patient has worn an infusion set for two days, the historical physiological glucose data and / or insulin delivery data processed by the regression model will be two days of data. Then, after an hour (when the model is run again), the data from the past hour is added to the final set of historical data, and the updated set of historical data (e.g., 2 days and 1 hour of data) is processed by the regression model.

[0060] The regression model outputs predictive data based on the aggregated physiological glucose data and insulin delivery data. In certain embodiments, processing the data using a regression algorithm includes calculating certain metrics (described herein) within successive one-hour time windows, then fitting a linear regression of the resulting sequence of metrics against time, and calculating a p-value representing the significance of the calculated time-linear coefficient.

[0061] Figure 4 provides an example of the calculation of the above-described metrics and fitting of linear regressions. Graph 120 in Figure 4 shows a plot of raw glucose measurements (e.g., glucose levels measured every 5 or 10 minutes) over two days, and graph 122 shows the calculated average glucose within successive 1-hour time windows. Graph 122 also includes a fitted linear regression 124 of the calculated average glucose.

[0062] In certain embodiments, the forecast data includes regression coefficient p-values ​​using selected metrics calculated over specific time intervals (e.g., one-hour intervals). The selected metrics may include a combination of two or more of the following:

[0063] [Table 1]

[0064] To reduce the complexity and computing resources required to run the regression algorithm and machine learning model (discussed further below), a subset of the metrics listed above may be selected. The subset of metrics may be selected based on how much independent impact a given metric has on the machine learning model. As an example, if two metrics have a high correlation (suggesting redundancy), only one of the metrics may be selected. As another example, if a given metric has a low contribution (or impact) to the output of the machine learning model, that metric is not selected. As a specific example, if a given metric has a Shaply value below a predetermined amount (e.g., less than 0.25), that metric is not selected.

[0065] Additionally, data other than that generated by a regression algorithm may be selected as features for input into a trained machine learning model, for example, the length of time that the current injection site has been in use may be a selected feature.

[0066] In a particular embodiment, the selected features, which are a combination of raw data and regression algorithm output, are those listed in the table below.

[0067] [Table 2]

[0068] Prediction data (e.g., p-values) are generated by the regression algorithm. In step 104, the data (e.g., attachment length) and prediction data are input into a trained machine learning model to generate an output, such as a prediction of infusion site failure.

[0069] The machine learning model is trained to receive data and prediction data as input, process the input, and then output a prediction of whether the injection site has failed. The trained machine learning model may comprise one of several different types of models, such as a neural network (e.g., a deep learning model), and may apply one of several different types of algorithms (e.g., a supervised algorithm such as a random forest or logistic regression). In certain embodiments, the trained machine learning model is an XGBoost model.

[0070] In certain embodiments, the machine learning model for a given patient is updated over time as the model receives patient-specific data. For example, a baseline machine learning model may be used when the patient uses their first few infusion sets (e.g., the first three months of infusion sets). However, after an initial number of infusion sets or an initial period, the machine learning model may be adjusted to address patient-specific trends. For example, a customized machine learning model for a given patient may be created by (1) determining (e.g., using average physiological glucose) which set of training data (e.g., 50-150 sets of training data) most closely resembles the patient's actual data over time, and (2) retraining the machine learning model based on the determined set of training data and the patient's actual data collected during the initial period. Additionally or alternatively, before the patient's actual data is used to retrain the machine learning model, the machine learning model must detect multiple infusion site malfunctions so that the actual data includes examples of infusion site malfunctions and data surrounding such detected malfunctions.

[0071] In certain embodiments, the output of the trained machine learning model is a probability (e.g., a value on a scale of 0-100%) that the infusion site has failed (step 106). This output can be used in a variety of ways to alert the patient and / or their healthcare provider. As one example, if the probability exceeds a certain value (e.g., 75%), a warning signal is generated and sent to the user and / or their healthcare provider. The warning signal can cause an alert to be displayed on the UI 38. As another example, the probability can be used to periodically update a window or status icon displayed on the UI 38. In this example, the window or status icon displays the current status of the infusion set (e.g., within the past hour or since the last output). This can include displaying a numeric value for the probability on the UI 38, displaying a graphic such as a battery indicator representing the remaining capacity of the infusion set, displaying a traffic light graphic responsive to the probability value, or simply displaying a different colored icon indicating the probability value. For example, if the probability is less than 50%, the window or status icon may indicate that the infusion set is operating properly (e.g., by displaying a particular color, phrase, text, number, or icon). If the probability is between 50% and 75%, the window or status icon may indicate that the infusion set may soon need to be replaced. And, if the probability is greater than 75%, the window or status icon may indicate that the infusion set needs to be changed, and a separate warning may be generated and displayed (e.g., in a pop-up window). The various numerical ranges discussed immediately above may be modified as needed.

[0072] The model-based approach described above may be executed on a server (e.g., in the cloud), on a UI38 (e.g., via an application on the patient's smartphone), on the drug delivery device itself (e.g., via an insulin pump), or by a combination of system components.

[0073] Rule-based approach The second disclosed approach to detecting infusion site malfunction is referred to in this description as a rule-based approach. Briefly, the rule-based approach applies a set of rules designed to determine whether blood glucose data indicating hyperglycemia is caused by an infusion site malfunction or another cause, such as a missed or low meal bolus.

[0074] FIG. 5 outlines a method 150 for detecting infusion site malfunction using a rules-based approach. In step 152, baseline patient statistics are calculated. These baseline statistics may be based on the patient's physiological glucose data and insulin delivery data. For example, the baseline statistics may include average physiological glucose, TAR (time above range), and nadir and peak glucose during the baseline period. In certain embodiments, if a new infusion set has been in place for less than three days, the baseline period is initially the first two days. The baseline period may then be extended to the first three days if the new infusion set has been in place for more than three days. In certain embodiments, method 150 begins only after the patient has been in place for at least two days.

[0075] The baseline statistics are compared to the patient's most recent physiological glucose and insulin delivery data in step 154. In certain embodiments, the most recent data is data collected within the past 36 hours, even if this period overlaps with the baseline period.

[0076] In step 156, in certain embodiments, the baseline statistics are (1) first compared to thresholds established to detect whether a chronic infusion site failure has occurred, (2) then compared to thresholds established to detect whether an acute infusion site failure has occurred, and (3) then compared to thresholds established to detect whether an abnormal deviation has occurred. Various thresholds, described in more detail herein with respect to Figures 6-8, are established to detect risk of hyperglycemia.

[0077] Figures 6-8 provide additional details of detecting infusion site malfunction using a rules-based approach as outlined in Figure 5. Method 200 of Figure 6 includes inputting physiological glucose data and insulin delivery data into a computing device, such as controller 44 (step 202). In step 204, the data is first analyzed to determine whether a chronic infusion site malfunction may have occurred. Steps 202 and 204 of method 200 of Figure 6 correspond to steps 152 and 154 of method 150 of Figure 5. The remaining description in this section provides further examples of step 156 of method 150 of Figure 5.

[0078] If the data suggests a chronic infusion site malfunction, the data is further examined to distinguish between an infusion site malfunction or another cause of a chronic hyperglycemic event (step 206). Based on certain criteria, described further below, the logic will determine whether a chronic infusion set malfunction has occurred (labeled "red" in FIG. 6), is likely to occur (labeled "yellow" in FIG. 6), or should be ignored (labeled "green" in FIG. 6). If the data does not suggest a chronic site malfunction, the data is then analyzed in step 208 to determine whether an acute infusion site malfunction has occurred. If the data suggests an acute infusion site malfunction, the data is further examined to distinguish between an infusion site malfunction or another cause of an acute hyperglycemic event (step 210). Based on certain criteria, described further below, the logic will determine whether an acute infusion set malfunction has occurred, is likely to occur, or should be ignored. Finally, if the data does not suggest an acute site malfunction, the data is analyzed in step 212 to determine whether an abnormal excursion associated with an infusion site malfunction has occurred. The logic may dictate that an infusion set malfunction has likely occurred or that the deviation should be ignored.

[0079] Chronic infusion site failure includes situations in which insulin delivery becomes less effective over time at an infusion site. Therefore, the threshold for this first comparison to the baseline statistical value is set to detect whether physiological glucose has increased to a degree that suggests a risk of hyperglycemia. In the case of chronic infusion site failure, step 204 includes determining (1) whether the patient's TAR is higher than a threshold value (e.g., 25%) and (2) whether the baseline TAR has increased by more than a threshold value (e.g., 30%). Step 204 may also include determining (3) whether the nadir physiological glucose has increased by more than a threshold value (e.g., 70%) compared to the baseline statistical value or whether the mean physiological glucose has increased by more than a threshold value (e.g., 30%) from the baseline statistical value. If these various thresholds are violated, it suggests a risk of hyperglycemia. However, additional criteria may be applied to determine whether the risk of hyperglycemia is due to a temporary increase in physiological glucose or to chronic infusion site failure.

[0080] 7 outlines logic 250 for distinguishing between a temporary increase in physiological glucose and a chronic problem. At 252, the logic includes comparing insulin delivery amounts from a baseline period and a recent period. In one specific embodiment, this comparison is performed using a Wilcoxon rank-sum test. At 254, if the comparison indicates that the recent period included a greater amount of delivered insulin, the logic has detected a chronic insulin setting malfunction.

[0081] If the comparison at 256 indicates that the most recent period included a similar amount of delivered insulin, a second comparison is made at 258. This comparison involves determining (1) whether the peak physiologic glucose for the most recent period is above a threshold value (e.g., 250 mg / dL) and (2) whether the peak physiologic glucose for the most recent period increased by a threshold value (e.g., 30%) compared to the baseline statistical value. If both criteria at 258 are met, the logic indicates that a chronic insulin setting failure has been detected. If not, the logic indicates that there is still some risk (but a lower risk) of chronic infusion site failure of hyperglycemia.

[0082] If the comparison at 260 indicates that the most recent period included a smaller amount of delivered insulin, a second comparison is made at 262. This comparison involves determining whether the peak physiologic glucose for the most recent period was above or below a threshold (e.g., 250 mg / dL). If the peak physiologic glucose is above the threshold, then by logic there is some risk (but a lower risk) of hyperglycemia due to chronic infusion site failure.

[0083] As described above, after the baseline statistics are compared to thresholds established to detect potential chronic infusion site malfunction (as described in the immediately preceding paragraph), the baseline statistics are compared to thresholds established to detect potential acute infusion site malfunction. Acute infusion site malfunction includes situations in which an infusion site malfunction causes a relatively dramatic increase in physiological glucose. In the case of acute infusion site malfunction, step 208 (of FIG. 6) involves determining whether physiological glucose has increased above a threshold (e.g., 180 mg / dL) starting from a lower threshold (e.g., 70 mg / dL) over a specific time period (e.g., 7 hours). If such criteria are met, it suggests a risk of hyperglycemia. However, additional criteria may be applied to determine whether the risk of hyperglycemia is due to reasons other than acute insulin site malfunction.

[0084] 8 outlines logic 300 for determining whether a dramatic increase in physiological glucose is the result of acute insulin site failure or another reason. In short, if the most recent insulin bolus is greater than that delivered during the baseline period, the sudden increase in physiological glucose is likely due to infusion site failure rather than a missed bolus or small meal bolus. In certain embodiments, for acute failure analysis, the baseline period is shorter (e.g., 12 hours) than for chronic failure analysis.

[0085] At 302, the logic includes determining the start and end times of the determined physiological glucose surge. At 304, the amount of the previous maximum effective bolus is determined. In certain embodiments, the previous maximum effective bolus is calculated as the maximum bolus taken over a set amount of time (e.g., 6 hours) before the start and / or end times of the determined physiological glucose surge. Also at 304, the previous maximum effective bolus is compared to the median bolus from the baseline period. At 306, the peak physiological glucose for the most recent period is compared to that of the baseline statistics. If the comparison indicates that the most recent peak is greater than a threshold (e.g., 30%) compared to the baseline, the logic has detected an acute insulin setting failure. If not, the logic still indicates some risk (but a lower risk) of hyperglycemia due to an acute infusion site failure. At 308, the peak physiological glucose for the most recent period is compared to that of the baseline statistics. If the comparison shows that the most recent peak is greater than a threshold (e.g., 30%) compared to the baseline, then by logic there is some risk (but a lower risk) of hyperglycemia due to acute infusion site failure.

[0086] As described above, after the chronic and acute failure analyses, the rule-based approach may detect abnormal excursions. An excursion includes a situation in which physiological glucose increases from a nadir to a peak above a threshold (e.g., 180 mg / dL) and then decreases below the threshold until a nadir occurs. An excursion is abnormal if the physiological glucose and insulin delivery data indicate a decrease in control, such as when physiological glucose does not decrease rapidly after an insulin bolus is delivered. In an exemplary embodiment, detecting an abnormal excursion follows a multi-step process: (1) identifying the excursion period, (2) calculating the area under the curve (AUC) of the hyperglycemic region, and (3) analyzing the associated insulin delivery data.

[0087] For the first step, an excursion period is determined by identifying two nadirs and one peak in the physiological glucose data (e.g., a period starting with an increase in physiological glucose, a peak, and ending when physiological glucose stops decreasing). In certain embodiments, to help limit the effects of noise and / or small local excursions, the physiological glucose data is processed (e.g., via a local polynomial method) to smooth the physiological glucose data. An excursion is then considered not to have ended and combined under either of two circumstances: (1) the ending nadir is greater than 180 mg / dL; or (2) the ending nadir is between 70 mg / dL and 180 mg / dL and glucose remains below 180 mg / dL for more than two hours from the current peak to the next peak.

[0088] For each excursion period, there is a region that exceeds a certain threshold (e.g., 180 mg / dL) and can be referred to as the hyperglycemic region. The AUC for the hyperglycemic region can be calculated for both the baseline period and the most recent period (e.g., the last day). FIG. 9 shows an exemplary plot of physiological glucose data and the region that defines the AUC for the hyperglycemic region. The AUC is shown in FIG. 8 as a shaded region with a numerical indicator indicating the calculated area. If the AUC for any of the hyperglycemic regions during the most recent period is greater than the AUC for all of the baseline periods, the logic leads to further investigation.

[0089] Further investigations may include determining an indicator of an impaired physiological glucose response to insulin delivery. In certain embodiments, such an indicator is measured using a metric that measures the ability of insulin delivery to reduce physiological glucose, referred to herein as per unit insulin effect:

[0090]

number

[0091] If the per unit insulin effect during any excursion is statistically lower than the baseline, the insulin has a less effective ability to reduce physiological glucose. In certain embodiments, a threshold is set from the baseline distribution created by a bootstrap sampling approach to define whether an abnormal excursion is the result of an infusion site malfunction.

[0092] Combining model-based and rule-based approaches In particular embodiments, both model-based and rule-based approaches are utilized in various ways to complement each other.

[0093] As one example, two techniques are run simultaneously to provide a check on the other technique. For example, both techniques can be run periodically (e.g., approximately once every hour) and the results can be compared. If the rule-based technique determines that an event has occurred (such as the "red" or "yellow" event shown in the figure and described above), the output of the model-based technique can be used to confirm or modify the determined event. If the model-based technique outputs a prediction with a high level of confidence that contradicts the output of the rule-based technique, the output of the model-based technique can ultimately be used to determine the type of event. For example, if the rule-based technique results in a red or yellow event, the event is changed to a green event (e.g., no infusion site malfunction) if the probability of the event is less than a threshold (e.g., 5%). Conversely, if the rule-based technique results in a green event, the event is changed to red or yellow (e.g., indicating an infusion site malfunction) if the probability of the event is greater than a threshold (e.g., 95%).

[0094] As another example, a rule-based approach may be used alone until an event is detected. Once an event is detected, a model-based approach may be used to check the output of the rule-based approach. Because the rule-based approach requires fewer computing resources compared to the model-based approach, using this combination of approaches may reduce the overall power consumption of the system.

[0095] As another example, the model-based approach may be used exclusively for an initial period of wear (e.g., the first 2 or 3 days) because it is more sensitive to shorter periods of data. After the initial period expires, the system may switch to the rule-based approach, which is more stable with longer periods of data, until the infusion set is removed.

[0096] 10 depicts a method 400 for using both model-based and rule-based approaches. Method 400 includes determining the occurrence of chronic or acute infusion site failure based at least in part on a comparison of physiological glucose data and insulin delivery data to a threshold value (step 402). In certain embodiments, the comparison is between a threshold value and a calculated difference between baseline physiological glucose data and insulin delivery data and recent physiological glucose data and insulin delivery data.

[0097] In response to determining the occurrence of a chronic infusion site malfunction or an acute infusion site malfunction, the trained machine learning model may operate to confirm the occurrence of a chronic infusion site malfunction or an acute infusion site malfunction (step 404). For example, the trained machine learning model may output a value indicative of the likelihood of an infusion site malfunction. If the value is above and / or below a threshold, the occurrence of a chronic infusion site malfunction or an acute infusion site malfunction may be confirmed and an alert may be generated. In certain embodiments, the same physiological glucose data and insulin delivery data (or data derived therefrom) are used by both the rule-based approach and the model-based approach.

[0098] Further aspects of the present disclosure In certain aspects, as described above, a method utilizing a rules-based approach can be applied to determine the status of an infusion site, including calculating a baseline statistic associated with an initial time period and the infusion site, determining that a difference between the baseline statistic and physiological glucose data from a later time period exceeds a first threshold, and determining the occurrence of a chronic infusion site failure or an acute infusion site failure in response to determining that the difference between the baseline statistic and the physiological glucose data exceeds the threshold.

[0099] A further aspect of the method includes comparing the difference between the baseline statistical value and the insulin delivery data and determining that the physiological glucose data is above a second threshold.

[0100] In an additional aspect, the method includes determining that an aspect of the insulin delivery data is higher than a baseline statistical value, and in response, determining the occurrence of a chronic infusion site malfunction.

[0101] In an additional aspect, the method includes determining that an aspect of the insulin delivery data is higher than a baseline statistical value, and in response, determining the occurrence of an acute infusion site failure.

[0102] In other aspects, as discussed above, a method can be applied to determine an infusion site condition by combining one or more rule-based approaches described herein with one or more model-based approaches described herein. The method includes using a rule-based approach to determine the occurrence of a chronic infusion site malfunction or an acute infusion site malfunction based, at least in part, on a comparison of the physiological glucose data and insulin delivery data to a threshold value. The method further includes, in response to determining the occurrence of a chronic infusion site malfunction or an acute infusion site malfunction, confirming the occurrence of a chronic infusion site malfunction or an acute infusion site malfunction using a model-based approach that operates a trained machine learning model.

[0103] In additional aspects, the rule-based approach is performed more frequently than the model-based approach, for example, the rule-based approach may be used approximately once an hour, and the model-based approach may be used only after the rule-based approach determines the occurrence of a chronic or acute infusion site problem.

[0104] In an additional aspect, the rule-based approach and the model-based approach are performed simultaneously.

[0105] In additional aspects, the output of the model-based technique confirms or overrides the output of the rule-based technique, for example, if the output of the model-based technique indicates a high probability (e.g., 90% or greater, 95% or greater) of chronic infusion site failure or acute infusion site failure, the output of the model-based technique overrides the output of the rule-based technique.

[0106] In an additional aspect, the same physiological glucose data and insulin delivery data (or the same data derived therefrom) are used by both the rule-based approach and the model-based approach.

[0107] Accordingly, this disclosure is intended to embrace all such alternatives, modifications, and variations. Additionally, while several embodiments of the present disclosure have been illustrated in the drawings and / or discussed herein, the disclosure is not intended to be limited thereto, as the disclosure is to be broad as permitted by the art and it is intended that the specification be read in the same manner. Accordingly, the above description should not be construed as limiting, but merely as exemplifications of particular embodiments.

Claims

1. 1. A method for predicting an injection site condition, said method comprising: applying a regression model to the physiological glucose data and the insulin delivery data to generate predictive data; operating a trained machine learning model to process the prediction data and generate an output; and determining based on the output that the infusion site has failed or is likely to have failed.

2. generating a warning signal indicating that the infusion site has failed; The method of claim 1 , further comprising: in response to the warning signal, displaying a warning on a graphical user interface.

3. 3. The method of claim 1 or 2, wherein the physiological glucose data and insulin delivery data are aggregated over a period of time beginning when the infusion site is first used.

4. The method of claim 3 , further comprising calculating a linear regression based on the aggregated physiological glucose data and insulin delivery data, wherein the predicted data is based at least in part on the linear regression.

5. The method of any one of claims 1 to 4, wherein the physiological glucose data and the insulin delivery data are updated and aggregated according to a set schedule.

6. The method of any one of claims 1 to 5, wherein the output is a value indicative of a likelihood of infusion site failure.

7. The method of any one of claims 1 to 6, wherein the prediction data comprises p-values ​​of selected metrics.

8. The method of claim 7 , wherein the selected metrics include metrics selected from categories of metrics, the categories including physiological glucose variability and physiological glucose.

9. 9. The method of claim 7 or 8, wherein the selected metrics include time above range, low glucose index, total bolus, standard deviation, coefficient of variation, and lability index.

10. The method of any one of claims 1 to 9, wherein the trained machine learning model is an XGBoost model.

11. 11. The method of claim 1, wherein the trained machine learning model is customized for a patient by retraining the machine learning model using the patient's previous physiological glucose and insulin delivery data.

12. The method of any one of claims 1 to 11, further comprising updating a status icon associated with the injection site on a user interface based on the output.

13. A computer program product comprising instructions for causing one or more processors to carry out the steps of the method according to claims 1 to 12.

14. A computer readable medium having stored thereon the computer program product of claim 13.

15. A computer comprising the computer-readable medium of claim 14.

Images