Glucose control using digital twin metabolic models

Abstract

A diabetes management system optimizes glucoregulatory hormone delivery using a connected glucoregulatory hormone delivery system and a computing device with a processor and non-transitory memory. The system receives glucose management data (glucose, insulin delivery, heart rate, or carbohydrate consumption data) from sensors. A digital twin metabolic model is selected and initialized based on the received data. The system simulates diabetes treatment interventions to predict glucose dynamics and determines an optimized glucoregulatory hormone parameter (e.g., insulin dose or increment that minimizes hyperglycemia and hypoglycemia risk). The determined glucoregulatory hormone parameter is transmitted to the connected glucoregulatory hormone delivery system, which may include an automated insulin delivery (AID) system or a connected insulin pen system, for implementation. By integrating real-time physiological data with predictive metabolic modeling, the system enhances glucoregulatory hormone therapy, providing personalized adjustments to improve glucose control.

Description

Docket No. 503478.70111:(Tech ID 3305)GLUCOSE CONTROL USING DIGITAL TWIN METABOLIC MODELSRELATED APPLICATION

[0001] This application claims priority benefit of U. S. Provisional Patent Application No. 63 / 626,502 filed January 29, 2024, which is hereby incorporated by reference in its entirety.TECHNICAL FIELD

[0002] This disclosure relates generally to type 1 diabetes (T1D) or type 2 diabetes (T2D) treatment intervention for a person using glucoregulatory hormone therapy and, more particularly, to glucose control using digital twin metabolic models.BACKGROUND INFORMATION

[0003] The following paragraphs in this section are based on introductory material and general principles discussed in a scientific paper by Young G, Dodier R, Youssef JE, Castle JR, Wilson L, Riddell MC, and Jacobs PG, titled “Design and In Silico Evaluation of an Exercise Decision Support System Using Digital Twin Models” (J Diabetes Sci Technol. 2024 Mar;18(2):324-334. doi: 10.1177 / 19322968231223217. Epub 2024 Feb 23. PMID: 38390855; PMCID: PMC10973845). These passages have been adapted for inclusion here. Where references to specific named authors are explicitly cited below, other citations have been omitted for brevity. No admission is made that the referenced paper or any cited materials constitute prior art.

[0004] T1D is characterized by the autoimmune destruction of pancreatic beta cells, resulting in a lack of endogenous insulin production. People with T1D must administer exogenous insulin subcutaneously through multiple daily injections (MDI) or via insulin pumps.Continuous glucose monitors (CGMs) help individuals with T1D monitor their glucose levels in real time and alert users to dangerous glucose levels. Physicians and diabetes educators often recommend that patients adjust insulin doses based on CGM responses to carbohydrate intake and physical activity. However, these recommendations are generally provided only once every three months, and personalized guidance is rarely delivered more frequently. Maintaining stable glucose levels can be challenging, especially during spontaneous physical activity. While regular exercise improves long-term outcomes, it frequently leads to low glucose levels (glucose <70 mg / dL), causing acute complications. Effective strategies for managing glucose levels during exercise remain a need for individuals with T1D.Docket No. 503478.70111:(Tech ID 3305)

[0005] Managing insulin dosing and carbohydrate intake before, during, and after physical activity presents significant challenges. Prior studies have demonstrated that individuals with T1D exhibit highly variable responses to different forms of exercise. For example, resistancebased exercise often elicits a significant increase in catecholamine secretion, leading to a surge in hepatic glucose production. This increase in hepatic glucose production may overcompensate for the muscle glucose utilization rate during heavy resistance exercise, particularly when performed in a fasted state, potentially causing a rise in blood glucose levels. In contrast, aerobic exercise places a sustained high demand for glucose by peripheral tissues, which can surpass counterregulatory mechanisms, resulting in decreased blood glucose levels. Prolonged aerobic exercise frequently causes substantial drops in glucose levels, increasing the risk of hypoglycemia. As a result, a significant number of individuals with T1D avoid exercise due to concerns about low glucose levels.

[0006] A decision support tool is a software system designed to provide personalized recommendations to assist individuals with T1D in managing their glucose levels. These systems can integrate data from wearable sensors to offer insulin, carbohydrate, and exercise recommendations throughout the day. While several systems have been developed to assist individuals in adjusting basal insulin, calculating mealtime insulin boluses, or recommending carbohydrate intake to maintain safe glucose levels, few tools are available that provide exercise-specific treatment recommendations. Additional details regarding decision support systems are provided in Young GM, “Exercise Physiology in Type 1 Diabetes: Development of Metabolic Models and Decision Support Systems.” (Oregon Health & Science University; 2023; doi: 10.6083 / js956g59c.)

[0007] Clinical guidelines are frequently used to provide recommendations regarding exercise for individuals with T1D. These guidelines consist of heuristics established by experts, offering recommendations for actions to take before, during, or after anticipated exercise based on basic contextual data such as CGM readings or insulin dosing information. For example, Moser et al. propose recommendations such as consuming carbohydrates, adjusting insulin boluses, delaying exercise, or prematurely terminating exercise based on CGM levels. (Moser O., et al., “Glucose management for exercise using continuous glucose monitoring (CGM) and intermittently scanned CGM (isCGM) systems in type 1 diabetes: position statement of the European Association for the Study of Diabetes (EASD) and of the International Society for Pediatric and Adolescent Diabetes (ISPAD) endorsed by JDRF and supported by the AmericanDocket No. 503478.70111:(Tech ID 3305)Diabetes Association (ADA),” Diabetologia. Dec 2020;63(12):2501-2520. doi:10.1007 / s00125-020-05263-9.)

[0008] These guidelines are often difficult for patients to follow as they can be complicated. Some decision support systems have integrated these types of guidelines into smartphone apps to make it easier for patients to follow the recommendations. However, clinical-based guidelines may not be suitable for all people as they do not account for individual differences in insulin sensitivity, diet and exercise responses that can have a major impact on glucose levels during and following exercise.SUMMARY OF THE DISCLOSURE

[0009] Advancements in diabetes technology including accurate, nonadjunctive CGM and connected insulin pens and pumps, as well as automated insulin delivery (AID) and decision support systems (DSS) built on these devices, seek to alleviate user burden. AID systems in particular have demonstrated significant improvements in glucose outcomes, increasing glucose time in range (TIR, 70-180 mg / dL) and reducing both time above range (TAR, >180 mg / dL) and time below range (TBR, <70 mg / dL). Central to the development of AID and DSS have been virtual patient models, which are used as predictive models as well as the basis of in-silico simulation environments used for pre-clinical validation of the systems’ efficacy and safety. Unfortunately, approximating the human glucoregulatory system remains difficult and most virtual patient population simulators are lacking the individual-specific adjustments needed to ensure models and in-silico results translate to real-world individuals.

[0010] Digital twins are mathematical model representations of real -world phenomena. Digital twin-based tools can use personalized models to provide more individualized and effective treatment recommendations. Digital twins in diabetes can be designed using metabolic simulators which have been developed to model glucose dynamics in people with T1D. These simulators are typically designed using ordinary differential equations to model how carbohydrates, insulin, and other hormones impact glucose dynamics in metabolism. Data-driven digital twins have been used in combination with coaching and medication interventions to help reduce HbAlc in people with type 2 diabetes.

[0011] Disclosed is a decision support system (DSS) based on ODE-based metabolic digital twin models that may be used to provide personalized treatment recommendations prior to either aerobic or resistance exercise (i.e., an exDSS) or on a periodic (e.g., weekly) basis for aDocket No. 503478.70111:(Tech ID 3305)carbohydrate-to-insulin ratio (CR), correction factor (CF) and long-acting insulin (i.e., a periodic DSS). For example, by integrating data from CGMs, insulin pumps, meal logs, and wrist-mounted heart rate monitors, the exDSS is designed to provide safe and effective insulin, carbohydrate, and behavioral recommendations prior to exercise to help people with T1D or T2D exercise more safely and avoid low glucose (<70 mg / dL). As described herein, the exDSS improves glucose outcomes for an in silico population of people with T1D or T2D by reducing time in low glucose during and following exercise compared with (1) no intervention and (2) consensus clinical guidelines around exercise.

[0012] In one aspect, a diabetes management system for optimizing glucoregulatory hormone delivery includes a connected glucoregulatory hormone delivery system and a computing device comprising a processor and a non-transitory computer-readable medium storing instructions. When executed, these instructions configure the diabetes management system to receive physiological glucose management data (e.g., glucose, insulin delivery, heart rate, and carbohydrate consumption data) from the connected glucoregulatory hormone delivery system or associated sensors. The system selects a digital twin metabolic model, initializes its states, and simulates diabetes treatment interventions to evaluate predicted glucose dynamics. Based on an intervention yielding an extremum calculated diabetes treatment benefit, the system determines a glucoregulatory hormone dosage parameter and transmits it to the connected glucoregulatory hormone delivery system for delivery.

[0013] The system may further include a wearable fitness device providing heart rate data, enabling glucoregulatory hormone adjustments based on predicted glucose changes due to physical activity. The connected glucoregulatory hormone delivery system may include an insulin pump or a connected insulin pen system. Associated sensors may include a CGM and an insulin relay personal diabetes manager (PDM) or smart device. The system may further be configured to select insulin doses or adjustments that minimize hypoglycemia or hyperglycemia risk. Adjustments may be determined before, during, or after exercise based on heart rate data and carbohydrate intake and may be applied periodically for long-term glucose control. The system may also continuously update the digital twin metabolic model using real-time glucose and insulin data to provide predictive insulin delivery adjustments.

[0014] In another aspect, a method for optimizing glucoregulatory hormone delivery is performed by a diabetes management system including a computing device configured with a diabetes treatment application. The method includes receiving physiological data from aDocket No. 503478.70111:(Tech ID 3305)connected glucoregulatory hormone delivery system and associated sensors, including glucose levels from a CGM, insulin delivery data from a connected glucoregulatory hormone delivery system, heart rate data, and carbohydrate consumption data. In response to a scheduled event or an on-demand request, a digital twin metabolic model is selected and initialized based on the received data. The method further includes simulating diabetes treatment interventions, calculating diabetes treatment benefits, and evaluating predicted glucose dynamics. A glucoregulatory hormone dosage parameter is determined based on an intervention yielding an extremum calculated diabetes treatment benefit, and the parameter is transmitted to the connected glucoregulatory hormone delivery system for delivery.

[0015] The personalized metabolic model of glucose dynamics may include a set of ordinary differential equations (ODEs). The connected glucoregulatory hormone delivery system may include an AID system with an insulin pump or a connected insulin pen system with a smart insulin pen. The calculated diabetes treatment benefit may be based on a sum of the low blood glucose index (LBGI) and high blood glucose index (HBGI), Ambulatory Glucose Profile metrics, or a CR adjustment score. The calculated benefit may also correspond to a CF adjustment score or a combined CR and CF adjustment score based on the difference between minimum and baseline glucose.

[0016] The method may further include diabetes treatment interventions such as modifying exercise duration, intensity, or timing, adjusting carbohydrate intake, or delaying exercise. Other interventions may include modifying CR for short-acting insulin calculations, basal insulin infusion rates, CFs for high glucose, or long-acting insulin doses. The method may also provide periodic recommendations for adjusting CR, CF, and long-acting insulin. The insulin pump may autonomously adjust insulin delivery based on the received insulin dose or adjustment, while a connected insulin pen system may display the insulin dose or adjustment for manual confirmation before injection.

[0017] Additional aspects and advantages will be apparent from the following detailed description of embodiments, which proceeds with reference to the accompanying drawings.BRIEF DESCRIPTION OF THE DRAWINGS

[0018] To easily identify the discussion of any particular element or act, the most significant digit or digits in a reference number refer to the figure number in which that element is first introduced.Docket No. 503478.70111:(Tech ID 3305)

[0019] FIG. 1 is a block diagram of a decision support system in accordance with one embodiment.

[0020] FIG. 2 is a block diagram of a metabolic model in accordance with one embodiment.

[0021] FIG. 3 is a timing diagram comparing performance of digital twins simulating glucose in response to insulin, carbohydrates, and heart rate in accordance with one embodiment.

[0022] FIG. 4 are a pair of mean error diagrams showing adaptability of digital twins in accordance with one embodiment.

[0023] FIG. 5 is a table showing treatment interventions in accordance with one embodiment.

[0024] FIG. 6 is a pair of tables showing reduced hypoglycemia and improved time in range for aerobic and resistance exercise in accordance with one embodiment.

[0025] FIG. 7 is a pair of tables showing improved glucose outcomes compared with no intervention and with consensus guidelines for aerobic and resistance exercise in accordance with one embodiment.

[0026] FIG. 8 is a table showing expected benefit of CR and CF candidate recommendations in accordance with one embodiment.

[0027] FIG. 9 is a block diagram of a system, according to one embodiment.

[0028] FIG. 10 is a block diagram of a computing device, according to one embodiment.

[0029] FIG. 11 is a flow diagram of a process in accordance with one embodiment.DETAILED DESCRIPTION OF EMBODIMENTS

[0030] The following paragraphs are also based on details and analysis described in the aforementioned scientific paper by Young G., et al., titled “Design and In Silico Evaluation of an Exercise Decision Support System Using Digital Twin Models.” As noted previously, the passages have been adapted for inclusion here, with some citations omitted for brevity, and no admission of prior art.

[0031] FIG. 1 shows a diabetes management system 100 including a personalized computing device 102 configured (e.g., using a smartphone app 104) to select an optimal diabetes treatment intervention, according to personalized metabolic modeling of glucose dynamics 106. As described in this disclosure, diabetes management system 100 selects the optimal T1D or T2D diabetes treatment intervention by determining an optimal digital twin at a given time, initializing its states, and simulating a set of available interventions to compare a benefit ofDocket No. 503478.70111:(Tech ID 3305)each intervention. In some embodiments, the optimal intervention is then determined as the one that yields the best sum of the LBGI and the HBGI.

[0032] Computing device 102 may include a mobile phone, insulin pump, or server. In the example of FIG. 1, computing device 102 is a smart phone configured to communicate with wearable glucose sensing and regulating medical devices 108. These include an Insulet Omnipod insulin pump 110 and a Dexcom G6 CGM 112 (or similar device configured to implement features for glucose monitoring). At least a part of wearable glucose sensing and regulating medical devices 108 may be an AID system, which dynamically adjusts insulin delivery in response to real-time glucose levels, leveraging closed-loop control algorithms to optimize glucose regulation. This integration allows diabetes management system 100 to provide more precise and adaptive glucose control. Furthermore, wearable glucose sensing and regulating medical devices 108 may be an automated (multi)hormone delivery (AHD) system for delivery of one or more of insulin, glucagon, or pramlintide.

[0033] Also shown is an M-600 fitness watch 114 for quantifying exercise (heart rate).Wearable glucose sensing and regulating medical devices 108 also include an Insulet insulin relay PDM 116. Data is stored to an Amazon Web Services (AWS) data repository 118 (e.g., AWS Cloud monitoring and data storage).

[0034] This disclosure initially describes how diabetes management system 100 provides an exercise-related diabetes treatment intervention shown and described with reference to FIG. 2-FIG. 7. In some embodiments, exercise-related diabetes treatment interventions are selected on demand, e.g., either 90-minutes prior to exercise based on meal adjustments (IM) or they are done immediately prior to the start of the exercise (IE). Later, this disclosure also describes how diabetes management system 100 acts as a DSS to provide recommendations for changes to CRs, CFs, and long-acting insulin amounts or adjust glucoregulatory hormone dosage parameters. Glucoregulatory hormone dosage parameters include the actual hormone dosage amount, an incremental adjustment to the dose, a temporal adjustment to the timing of administration, or other modifications to dose-related information.Exercise-aware metabolic model and digital twins for forecasting glucose dynamics

[0035] FIG. 2 shows an example of a metabolic model 200 for forecasting glucose dynamics. The model is an extension of an open source simulation metabolic model published by Resalat et al. (Resalat N, El Youssef I, Tyler N, Castle J, Jacobs PG. A statistical virtual patientDocket No. 503478.70111:(Tech ID 3305)population for the glucoregulatory system in type 1 diabetes with integrated exercise model. PLOS ONE. 2019;14(7):e0217301. doi: 10.1371 / journal.pone.0217301.)

[0036] FIG. 2 shows metabolic model 200 has model inputs (shown in blue) including subcutaneous insulin, oral carbohydrates, and heart rate (HR). Glucose storage compartments are represented by circles for carbohydrate absorption (Mi and M2) and for disposal of glucose (Qi, and Q2), with Qi representing disposal out of plasma. Storage and action compartments for insulin kinetics (Si, S2, 1) and insulin dynamics (Xi, X2, and X3) are represented by circles. An exercise-aware compartment Ai takes HR as an input to impact both endogenous glucose production (EGP) and glucose disposal glucose disposal out of compartment Qi. Fluxes between compartments are denoted by arrows, with solid arrows signifying direct mass transport and dashed arrows representing indirect effects. Disposal fluxes are represented by downward-pointing arrows.

[0037] Metabolic model 200 represents ordinary differential equations developed using physiology tracer study data from people with T1D in an aerobic exercise study or a resistance exercise study.

[0038] In these two physiology studies, 25 participants in each study completed in-clinic aerobic and resistance exercises while glucose levels were held constant by infusing dextrose combined with a glucose tracer (di -deuterated glucose) using a euglycemic clamp. This allowed for close approximation of endogenous glucose production and glucose disposal during and following the different types of exercise. Candidate ordinary differential equation models of glucose metabolism during exercise were proposed based on these data sets. Model parameters were identified using Bayesian sampling using Stan (S. D. Team, in Stan Modeling Language Users Guide and Reference Manual, 2.26.0 ed, 2020) on data collected from participants in the physiology study during their in-clinic aerobic and resistance exercise sessions. Model equations and parameters are as follows.Equation Set 1Q̇₁ = -k₁₂Q₁ + EGP₀(1 - X₃) - (X₁ + k₀₁)Q₁ + K • M₂Q̇₂ = x₁Q₁ - (k₂ + k₁₂)Q₂

[0039] In Equation Set 1,and Q2represent the first derivatives of the glucose compartments Qi and Q2, respectively. The parameter kl2is a fixed parameter that describesDocket No. 503478.70111:(Tech ID 3305)the glucose flux from compartment Q2into Q Glucose flux from (f into Q2, glucose utilization from the compartment Q2, and EGP are affected by the corresponding insulin action compartments, X1, X2, and X3, respectively. EGP0is a fixed parameter describing the basal rate of endogenous glucose production. Parameter K is a fixed delay parameter used to model inputs of oral carbohydrates via glucose compartment M2using the participants’ body mass, BM [kg]. Insulin independent glucose utilization is modeled by the time varying parameter, kQ1[min-1], and controls the speed of glucose disposal from the observable compartment. This parameter changes based on the concentration of glucose inand is assumed to be fixed when plasma glucose levels are within a glucose range of 4.5-9 mmol / L. In this work on the exDSS, Agis considered to be 1.0. Agis defined as the active grams of carbohydrates absorbed in the blood when a carbohydrate is consumed. Vgrepresents the volume of distribution of glucose in the accessible compartment Qj.

[0040] Equation Set 2 shows how the k01parameter changes with the level of glucose stored in the observable compartment. G is the glucose amount in blood.Equation Set 2Fc01mmolif G < 4.5 — —(4.5) - I / / LtFc01mmolfcOi— ■* f 4.5 < G < 9 — —1 / 4LtF01Vnmmol0.003 + -j- - 0.003 • (9)Qi * Qi —wr hieTG r G - — Ql / fand J K v - —51556' A[g

[0041] Equation Set 3 models the oral carbohydrate subsystem of the glucoregulatory model. The model accounts for the delayed kinetics of oral carbohydrate intake as food is digested and absorbed into circulation.[grams] and M2[grams] represent storage compartments for meals, with Mxdraining directly into M2. Carbohydrates stored in M2are transferred into the observable glucose compartment (f via a bioavailability parameter, K. Carbohydrate disposal from this system is modeled as a flux out of compartment M2. Ug[grams] represents oral carbohydrate intake rate and tmax, G is the delay parameter for the oral carbohydrate subsystem.Docket No. 503478.70111:(Tech ID 3305)Equation Set 3Ṁ₁ = Ug / tmax,G - M₁ / tmax,G

[0042] Equation Set 4 describes the subcutaneous insulin kinetics subsystem of the glucoregulatory model, which governs insulin injection, absorption, and distribution in the body. Compartments [mU / kg] and S2[mU / kg] are subcutaneous insulin kinetics compartments for insulin, with releasing insulin to S2at a rate determinedby max- I represents the active insulin in circulation and is directly fed from compartment S2Active insulin is eliminated from compartment I via the disposalparameter, ke. U] [mU / kg / min] represents the subcutaneous insulin injection rate, tmaxis the kinetics delay parameter, and V, is the volume of distribution for insulin.Equation Set 4Ṡ₁ = (Uᵢ - S₁ / tmax)Ṡ₂ = S₁ / tmax - S₂ / tmaxİ = S₂ / (Vᵢ · tmax) - keI

[0043] Insulin exerts its effect on the overall glucoregulatory model via the insulin action compartments Xj [min-1], X2[min-1], and X3[unitless], as described in Equation Set 5.Circulating insulin in compartment I contributes to each of the action compartments via the inactivation rates kal, ka2, and ka3, as well as the personalized insulin sensitivityfactors sf₁, sf₂, and sf₃. Insulin stored in each of the activity compartments decreases with time according to the rate constants kₐ₁, kₐ₂, and kₐ₃.Docket No. 503478.70111:(Tech ID 3305)Equation Set 5= sfl^all ~ ^al^l%2= Sf 2^0.2! ~ ^a2^2%3 ~Sf 3^0.3! ~ ^0.3% 3

[0044] The model input indicating the intensity of the exercise is the change in heart rate relative to a baseline heart rate (A / / / ?5min(t)) whereby the baseline heart rate is estimated during a sedentary period. The heart rate data were measured via a Verily wrist mounted activity tracker (Google). The raw heart rate measurements were converted to a delta heart rate according to Equation Set 6, and this delta heart rate metric was used to inform the exercise model of physical exertion that then impacts glucose disposal as shown in FIG. 2. The participants’ baseline heart rate HRbaseline[bpm] (taken during a sedentary time) was subtracted from each of the n recorded heart rate measurements, HRi(t) [bpm] from the previous five minutes and clipped to a minimum value of zero. The clipping ensured that the exercise metric was always non-negative and trended towards zero during sedentary periods when participants’ measured heart rate was close to their baseline value. The final activity metric, AH / ?5min(t) [bpm], was calculated by taking an average over all the recorded heart rate measurements during the past five minutes (HRit')').Equation Set 6: Delta Heart Rate Activity Metricn.A r l r> \ ' (HRi \t) HRbaseiineAH / ?5min(t) = ) max -, 0Z— 1 \ n / i=0x z

[0045] During model fitting, HRbaselinewas calculated for each participant using an average of the measured heart rate during the final 30 minutes of cool-down for each in-clinic exercise session, which was 223 minutes after exercise ended, on average. This model was validated in terms of its accuracy in estimating glucose during real-world exercise using the TIDexi data set. During model evaluation using the TIDexi study dataset, HRbaselinewas chosen for each participant using their average resting heart rate collected during study onboarding when the participants were not exercising and were resting.

[0046] The exercise model in FIG. 2 includes both instantaneous and delayed effects of elevated heart rate on glucose utilization. Hepatic EGP is affected by the delta 5-minute heartDocket No. 503478.70111:(Tech ID 3305)rate feature (AH / ?5min(t)) as well as the insulin action compartment Xt. Exercise can exert its effect in the model through instantaneous as well as delayed mechanisms. The delta heart rate activity metric, A / U?5min(t), is used to directly impact increased glucose utilization through interaction with the parameter k01via the fitted parameter a. Model Q incorporates an activity delay compartment, A1, that is controlled A / f / ?5min(t). Activity effects stored in this compartment impact k01through the fitted exercise parameter kAand are slowly drained according to the elimination rate parameter, kQA. Equation Set 7 shows the entire activity subsystem for proposed model Q.Equation Set 7fcOi= kAAr+ aAH / ?5min(t) + / ?EGP =YEHRSmin(f) + EGPO(1 - X3)A=A / 7 / ?5min(O—

[0047] Personalized exercise model parameters were fit for each participant in a physiology tracer study model fit study dataset. Using the physiology data from the two NIH-funded tracer studies, digital twins were generated for the aerobic exercise models and another set of digital twins for the resistance exercise models. These digital twins had unique parameters for the exercise portion of the model (e.g., each twin had a unique value of a,, kA, koA, EGPo, and y in the model in FIG. 2). These personalized models use insulin, carbohydrates, and heart rate as inputs and then forecast glucose. Exercise parameter values were constrained to fall within the ranges selected based on prior studies and physiological plausibility.

[0048] For every aerobic and resistance exercise video session that participants did during the TIDexi study, a digital twin was matched to determine which twin was the best metabolic match to that exercise event. When the exercise parameters of the model were fit, the insulin sensitivity model parameters,Sf2, and Sy3,were held constant at their mean population values for all study participants. During model evaluation, immediately prior to the exercise event, individual TIDexi study participants were matched to one of 99 digital insulin sensitivity twins based on which twin’s simulated CGM achieved the closest match to the measured TIDexi study CGM during the pre-exercise period. These sx, Sf2, and Sf3parameters were used to initialize the conditions of the model and ensure that all model states start at the proper initial value.Docket No. 503478.70111:(Tech ID 3305)

[0049] Models were run-in to 90 minutes prior to exercise (t = -90) rather than the start of exercise (t = 0). This allowed for simulation of pre-exercise interventions such as carbohydrate consumption or adjustment of meal insulin dosing in anticipation of exercise.

[0050] Oral carbohydrate and insulin data from each scenario were given as inputs to the metabolic model four hours prior to the start of exercise up to 180 minutes prior to the start of exercise (t = -40 to t = -180). This enabled estimation of the initial states of the oral carbohydrate and insulin kinetics model subsystems at minute t = -180. Next, all 99 insulin sensitivity twins were simulated using Model Q whereby exercise was considered ‘off’ by supplying a delta heart rate of zero to the model. The optimal insulin sensitivity parameters were selected (s^. Sf2, and Sy3) in Model Q by determining which of the digital twins matched the real-world data most closely, as determined by having the minimum root mean squared error (RMSE). The twin with the closest estimated glucose to observed CGM values (via RMSE) was selected as that session's insulin sensitivity twin, and the twin's values of s -1;s2, and Sf3were used for all subsequent simulations of that exercise session. During the identification of S, Sf2, and s^3, the simulator was free to select whichever set of s^, Sf2, and Sf3parameters achieved the closest match to observed data.

[0051] Using the insulin sensitivity digital twin parameters selected during run-in, it was determined which set of exercise model parameters, or exercise digital twin, best described each individual's unique and dynamic physiology during the exercise session drawn from the TIDexi study dataset. For each aerobic or resistance TIDexi study exercise session, every exercise digital twin was simulated from either the aerobic exercise twin cohort for aerobic exercise, or the resistance exercise twin cohort for resistance exercise, respectively. The exercise digital twin that produced estimated glucose closest to the observed CGM as determined by minimum RMSE across the exercise and post-exercise period was selected as the matching exercise digital twin, and its set of six exercise parameters was used for the remainder of simulations for that given exercise scenario. The initial states of the exercise digital twin were identified using the input data (insulin, carbohydrates, and heart rate) for that exercise session, from minutes t = -240 to t = -90, to estimate Model Q’s insulin, carbohydrate, and activity subsystem compartment states ({XltX2, X3, AltMltM2, S1, S2, 1t= —90) at the start of the pre-exercise period. In other words, each digital twin was simulated just for the exercise portion to find the match, then the process went back and initialized that matching twin’s states pre-exercise. Then, the various intervention scenarios were ran.Docket No. 503478.70111:(Tech ID 3305)

[0052] Once all model parameters and pre-exercise subsystem model initial states had been estimated using the described run-in method, the remaining unknown model states, Qi and Q2were estimated using recorded CGM data from t = -100 to t = -90. Q2was estimated using Equation Set 8 and the mean CGM rate of change from a three-point linear regression across the 10 minutes of CGM data recorded immediately prior to the start of exercise.Equation Set 8(k01+ Xr) - Q1- EGP - K - M1- Q1Q~ ^12

[0053] Qi was set equal to the recorded CGM at minute t = -90 and all recorded real-world insulin and CGM data after t = -90 were ignored. Starting from t > -90, parallel simulations of each available exDSS intervention recommendations (see, e.g., FIG. 5) were ran following clinical guidelines, as described in Moser et al. Actual heart rate data recorded during each TIDexi study exercise session was used for each simulation.

[0054] All simulated participants received a constant basal insulin infusion, as calculated in Equation Set 9, across the entire pre-exercise, exercise, and post-exercise period (t = -90 to t = +90). This basal insulin amount was determined using Equation Set 9 below.Equation Set 9: Estimating Basal Insulin Infusion RateUlnsulinbasal= kb• TDIRhr.where kb= 0.6

[0055] Basal insulin infusion rates were estimated using the total daily insulin requirement (TDIR) from the insulin sensitivity twin that achieved the closest agreement to observed CGM. These insulin infusion rates were established in simulation by finding the constant insulin infusion rate that achieved a steady state estimated glucose of 130 mg / dL for each digital participant. Meal insulin bolus insulin amounts were estimated using a simple bolus calculator shown in Equation Set 10:Docket No. 503478.70111:(Tech ID 3305)Equation Set 10: Estimating Mealtime Insulin Bolus1 [I / ]Insulinbolus[U] = — • Insulinbasai— + Insulinmeal[U]IL Lnrl1Insulinmeai

[0017] Cctrbsmeai[g] • „ -CR [%]

[0056] Equation Sets 9 and 10 were used to calculate the amount of insulin delivered at every five-minute interval during the simulation. This amount included both basal insulin and meal insulin. The basal insulin was based on the person’s basal insulin rate that would keep them at a target of 130 mg / dL. The basal insulin rate (in units of units / hr) is divided by 12 since it is given once every five minutes (i.e. 12 boluses / hour). It was assumed that the initial instantaneous CR was a constant 10 g / U for all participants in this study.

[0057] An example timeline 300 of how well the digital twins matched one of the real-world TIDexi study participants is shown in FIG. 3. The results shown in FIG. 3 were generated by applying to the digital twins exercise event data from a TIDexi study dataset, which comprises CGM, ingested carbohydrates, heart rate, and insulin data collected at home. Input data from four hours prior to the start of recorded exercise up to one hour after the nominal cessation of exercise is applied to the digital twins to simulate glucose responses to exercise. In this example, a digital twin is fit during aerobic exercise (shaded green area). Top panel 302 is CGM during exercise. Blue trace is real-world CGM data 304 from a TIDexi participant; note that hypoglycemia 306 occurred post-exercise. Digital twin candidates 308 are light gray dashed. Population average 310 is shown by dashed red line. Best twin 312 (black dashed) is the twin most closely matching real -world CGM data 304 during the 90 minutes following a start of exercise 314. Bottom green trace shows how heart rate 316 increases during exercise and returns to baseline after the exercise event. Carbohydrate intake 318 is shown as red Xs with amount of carbohydrates in grams. Insulin infusion rates 320 are shown in purple.

[0058] Best-fitting digital twins were evaluated on 247 participants from the TIDexi data set who were randomized to perform either aerobic or resistance exercise during the study to identify digital twins and personalize the models. Mixed effects modeling assessed statistical differences between models. The model accuracy was calculated by the mean error determined by subtracting the model estimation of glucose during exercise from the actual glucose duringDocket No. 503478.70111:(Tech ID 3305)exercise in the TIDexi exercise sessions. It was found that personalized metabolic models were accurate in predicting glucose drops during aerobic and resistance exercise.

[0059] The model in FIG. 2 is designed to adapt over time as new exercise sessions are observed. Adaptation was done by identifying which digital twin would accurately forecast CGM values during exercise that best matched the person’s actual CGM (FIG. 3).

[0060] FIG. 4 shows that when participants in the TIDexi study performed their aerobic and resistance exercises, respectively, the model initially tended to overestimate the drop in glucose during exercise. However, after additional exercise sessions were observed and as the digital twin learned over time, the bias in the error was reduced and the best digital twin could more accurately predict glucose levels during exercise with reduced bias. A new digital twin was identified for each exercise session. Every time a new exercise session was done, model parameters from the digital twins identified from prior exercise sessions were averaged to create a personalized digital twin. Then the new personalized digital twin was used to forecast glucose dynamics for the new exercise session. For instance, after five exercise sessions have been done, it is possible to average the model parameters from each of the five digital twins to create a new personalized digital twin. This personalized digital twin will then be used in the sixth exercise session.

[0061] A left side of FIG. 4 shows accuracy in terms of mean error of the model in estimating glucose during and one hour after exercise for repeated TIDexi aerobic exercise video sessions. A right side of FIG. 4 shows accuracy during resistance exercise sessions. After the first two-three exercise sessions, the error bias is reduced to near zero demonstrating adaptability.

[0062] FIG. 5 shows a table summarizing examples of treatment interventions available in an exercise-aware decision support system. The exercise-aware decision support system uses the metabolic digital twin model in FIG. 2 to provide optimal recommendations to people prior to the start of exercise to improve glucose outcomes during and following exercise. Optimal recommendations were evaluated by simulating replays of the best digital twin across each possible recommendation. The possible recommendations included (1) changing the exercise duration, (2) changing the exercise intensity, (3) eating carbohydrates, (4) waiting to exercise, (5) adjusting CR in advance of exercise, or (6) adjusting basal insulin infusion in advance of exercise.

[0063] The exDSS was evaluated by simulating how multiple potential interventions impact glucose outcomes using the exercise-aware metabolic simulator. The exDSS providedDocket No. 503478.70111:(Tech ID 3305)recommendations to help the person maintain safe glucose during exercise and post-exercise periods. It was assumed that exDSS users would receive the exercise recommendation at least 90 minutes before initiating exercise. Exercise interventions could occur during any meal in the 90-minute pre-exercise period (ZM) or at the instant immediately prior to the start of exercise ( E). The exDSS provided recommendations to in silico study participants for any meals that occurred within the 90 minute pre-exercise period ( / M). This 90-minute window was estimated to be when users could reasonably anticipate future exercise and report the planned event to a smartphone-based DSS. The exDSS also provided recommendations immediately prior to the start of exercise ( / F).

[0064] For example, one simulated intervention may recommend eating 20g of carbohydrates immediately at the start of exercise, 1E. Another intervention may recommend increasing the CR by 30% at a pre-exercise meal, IM. One of the recommendations is to “change exercise intensity.” Changing the exercise intensity in the simulation involved scaling the heart rate by the factors listed in FIG. 5 under the column labeled Available Choices. Another intervention is to “change exercise duration.” Changing the exercise duration was done in simulation by adjusting the heart rate data during the exercise by either upsampling or downsampling the heart rate data to make the duration of the exercise session match the recommended exercise duration. The five-minute discretized heart rate data was generated via interpolation or upsampling of the recorded heart rate data.

[0065] The exDSS selected the best recommendation, ropt, by choosing the simulated intervention that yielded the best overall glucose control from the start of exercise up to one hour after the cessation of exercise, as quantified using optimal benefit. Optimal benefit was quantified as the sum of the LBGI and HBGI as has been done previously in Mosquera-Lopez et al. (C. Mosquera-Lopez, et al., “Predicting and Preventing Nocturnal Hypoglycemia in Type 1 Diabetes Using Big Data Analytics and Decision Theoretic Analysis,” Diabetes Technol Ther, vol. 22, pp. 801-811, Nov 2020.)

[0066] The LBGI and the HBGI penalize glucose values that fall outside of the target glucose range (70-180 mg / dL) and impose a higher penalty on values in the more dangerous low glucose range. After simulating all possible recommended interventions (7?) by replaying exercise sessions after the model initialization to time t = -90 before the start of exercise, the intervention that resulted in the optimal benefit (Equation Set 11), was selected as the optimal recommendation, ropt. This represents the best potential pre-exercise intervention that a userDocket No. 503478.70111:(Tech ID 3305)could take to maintain safe glucose levels and should be presented to the user before an anticipated exercise session via a decision support smartphone application. Note that the results describing the effect of exDSS on the optimal benefit in terms of glucose outcomes assume that only a single intervention, ropt, was taken before initiating exercise.Equation Set 11: Selecting Optimal Recommendations by Low and High Blood Glucose Indicesropt= min(LBGI + HBGI)rV r G Rn nwhere LBGI = -Y rl(CGMf), and HBGI = - Yn Z_i n i=l i=lf(CGM) = 1.794 ■ (In(CGM)1026- 1.861),10 • CGM2if f(CGM) < 00 otherwiserh^CGM} = (10•CGM2if f (CGW > 0(. 0 otherwise

[0067] The best recommendation to optimize glucose outcomes was selected by assessing LBGI summed with the HBGI. The LBGI and HBGI are risk-based indexes. The optimal intervention determined by exDSS was selected based on which recommended yielded the minimum sum of the HBGI and LBGI.

[0068] The sum of these two metrics was used to define the optimal benefit of the best recommendation when considering all recommendations (Equation Set 11). The sum of HBGI+LBGI penalizes recommendations that lead to glucose values that deviated too far from the target range, with a higher penalty assigned to low glucose values (<70 mg / dL) since these types of recommendations are considered the most dangerous. The asymmetric risk function for LBGI and HBGI serves to prevent recommendations from being selected as optimal if they lead to low glucose, which carries a greater risk of acute complications compared to the high glucose range (>180 mg / dL). The LBGI is higher when glucose is lower than 70 mg / dL while the HBGI is higher when glucose is higher than 180 mg / dL. Selecting interventions that minimize the sum of LBGI+HBGI will thereby minimize the time that a person spends outside of the target range of 70-180 mg / dL. In virtual participants, the benefit of the exDSSDocket No. 503478.70111:(Tech ID 3305)intervention (arm 1) was compared with consensus guideline recommendations (arm 2) and no intervention (arm 3).

[0069] The exDSS was compared with a no intervention arm and to an arm whereby the virtual patient followed clinical guidelines on best practices to follow prior to start of exercise. The Moser et al. guidelines were established by a consortium of clinicians and researchers with expertise managing glucose during exercise in people with T1D. The relevant interventions from these clinical guidelines are summarized in a series of tables and simulated participants selected treatment recommendations from these tables using their current CGM reading and CGM trend arrow information, which gives information about the current direction of glucose change and how fast glucose is currently changing. In the clinical guidelines arm of the study, in silico participants could take any the clinical guideline interventions before, during, or after exercise. The exercise recommendations described in Moser et al. were adapted for use in the OHSU simulator and are described in O. Moser, et al.

[0070] Note that while the Moser et al. clinical guidelines provided different CGM cutoff values for users with varying risk levels of low glucose, it was assumed a high risk level for all study participants and evaluated recommendations from the “EXO” risk group (defined by the clinical guidelines). The report also delineated different recommendations for two possible scenarios: an expected increase in CGM or an expected decrease in CGM, based on the type of anticipated exercise and whether the person was in the fasted vs. unfasted state. For this study, it was assumed that CGM was expected to decrease for all aerobic exercise sessions, whereas CGM was expected to increase for all resistance exercise sessions. It was further assumed that all virtual participants in the study received pre-exercise recommendations at the midpoint of the pre-exercise window (t = -45). Using the estimated CGM and trend arrow at t = -45, a recommendation was provided that may have included eating a carbohydrate immediately, taking a bolus of correction insulin according to Equation Set 12, and / or delaying the start of exercise. All recommended exercise delays assumed a constant delay of 20 minutes, and all performance metrics presented in the results of this study were adjusted to compensate for changes in the start and stop times of all exercise sessions.Docket No. 503478.70111:(Tech ID 3305)Equation Set 12: Estimating Correction Insulin Boluses / rrntn rmtjn 1 Insulincorrection[L / ] ( CG Mcurrenf. r CG Mtarget 1 • 1777(7 1at“L cf \ %£ ■ ( / ]mg 1700where CGMttnarrgnpett = 120——, and c Jf =TDIR „ -twin

[0071] During exercise, simulated participants also received recommendations from clinical guidelines according to O. Moser, et al. During exercise, virtual participants may have been instructed to consume a carbohydrate or take a correction insulin bolus immediately and then either continue exercise for the anticipated duration of exercise or halt exercise immediately. Recommendations were provided at the midpoint of exercise, which was 15 minutes after the start of the exercise session. Any prior recommendation that changed the start or stop times of the reported exercise session was taken into account when evaluating this intervention point such that the during exercise intervention was always done at the midpoint of the exercise.

[0072] Virtual participants could also receive a final, post-exercise carbohydrate recommendation via clinical guidelines. Virtual participants may have been told to consume a carbohydrate immediately if their CGM values were too low or expected to drop in the near future. Recommendations were provided at the midpoint of the exercise + post-exercise period, 15 minutes after the cessation of exercise. Any prior recommendations that changed the start or stop times of the exercise session were considered when evaluating this intervention point such that the end of exercise recommendation was always given at the end of exercise.

[0073] The exDSS was successful in reducing hypoglycemia and increasing time in range for both aerobic and resistance exercise as compared with either following consensus guidelines or following no recommendations (FIG. 6). FIG. 6 shows the exDSS reduced hypoglycemia and improved time in range for aerobic (left side of FIG. 6) and resistance exercise (right side of FIG. 6). Simulations run using real -world meal and exercise data from the TIDexi aerobic and resistance exercise sessions in combination with the updated OHSU T1D simulator with new models in FIG. 2. Statistical significance for exDSS and Clinical Guidelines are in comparison to “No Intervention”. (**: 0.01 > p > 0.001, ***: 0.001 > p > 0.0001, ****: p < 0.0001).Participants using exDSS had reduced probability of a low glucose during exercise by 59% for aerobic and 55% for resistance exercise compared with no intervention.Docket No. 503478.70111:(Tech ID 3305)

[0074] Use of the exDSS reduced the sum of LBGI and HBGI (FIG. 7). FIG. 7: exDSS significantly improved glucose outcomes compared with no intervention and with consensus guidelines for aerobic (A) and resistance exercise (B). The LBGI summed with the HBGI during and 90-minutes following exercise start was lowest for exDSS compared with no intervention and consensus guidelines.Periodic DSS

[0075] The DSS can also provide periodic (e.g. weekly) explainable recommendations for adjusting time-dependent CRs, CFs in the morning, afternoon, evening, and overnight. One set of possible recommender time frames is provided in the following table.Table 1: Recommender time frames used to provide decision support for CRs and CFs.Time nameTime rangeMorning 7 AM - 10:59 AMAfternoon 11 AM - 3:59 PMEvening 4 PM - 10:59 PMOvernight 11 PM - 6:59 AM

[0076] Long-acting insulin recommendations can also be presented by the DSS to the user periodically (e.g. weekly). Participants who take two basal doses each day may receive a recommendation for either the morning dose, evening dose, or both doses.

[0077] Digital twins based on ordinary differential equations or machine learning algorithms may be used to optimally select new recommended time-based CRs, CFs, and long-acting insulin amounts. For example, the ODE digital twins described above can be implemented by training a neural network model as described in the ODENet methods paper. The neural network implementation of the digital twins would behave similarly to how the ODE implementation of the digital twins behave. And the other methods used for selecting recommended time-based CRs, CFs, and long-acting insulin amounts would be the same as for the ODE implementation of the digital twins.CR and CF modification recommendations

[0078] CGM, short-acting insulin, long-acting insulin, meals, and exercise can be pulled from the database for each meal observation window within the recommender time frames. A meal observation window is the time period after a meal insulin dose has been taken. It is the time period when a digital twin can assess the person’s glucose response to the meal to determine if the correct CR has been used. A correction factor observation window is the time period after aDocket No. 503478.70111:(Tech ID 3305)correction insulin dose has been taken. It is the time period when a digital twin can assess the person’s glucose response to the correction insulin to determine if the correct CF has been used.

[0079] CR and CF observation windows can be identified from the recommender time frames given in Table 1. If a short-acting insulin dose is given within a given recommender time frame, it can then be grouped with that recommender time frame, even if the observation window ends in a subsequent time frame.

[0080] Example of how selection of optimal digital twin and selection of recommendations can be done: The digital twin can be selected from one of many candidate digital twins. For example, the digital twin could be selected from one of the 99 digital twins that are in the OHSU metabolic simulator.

[0081] Short-acting insulin doses may be selected from a given recommender time frame (e.g., select all short-acting doses in the morning) from one hour before the dose (to) until three hours after the dose (to- 1 to to+3). Digital twins may then be identified for each eligible shortacting insulin dose over the prior week. The observation window would typically be from the start of a short-acting insulin dose to up to three to four hours after that dose is taken depending on other insulin or exercise events that may take place in this window.

[0082] An observation window may be considered eligible for a possible recommended change to the CR or CF if the following criteria are met: (1) There is a meal insulin dosed as indicated by information from an insulin pump within ±45 minutes of a short-acting insulin dose. (2) The short-acting insulin dose (including meal and / or correction insulin) is within ±10% of the dose recommended by the smart bolus calculator indicating that the user adhered to the advice of the bolus calculator. A meal insulin dose could come from an app or from the pump. The user typically reports meal insulin using a meal insulin bolus using a meal bolus calculator for the pump.

[0083] Other considerations for special conditions include: if there are two short-acting insulin doses within ±1 hour of each other, they can be combined together and considered as a single dose; if multiple smart bolus calculator entries are within ±45 minutes of a short-acting insulin dose, then the smart bolus calculator dose that is closest in time to the actual shortacting insulin dose may be used; and there should be a minimum of at least two-hours for example, after the last short-acting insulin dose is recorded when no additional insulin or exercise events take place for the observation window to be eligible.Docket No. 503478.70111:(Tech ID 3305)

[0084] Determination of optimal twin: The optimal twin can be identified by (1) matching a steady-state digital twin to the one-hour steady-state period of time before a meal or correction dose was taken, and (2) matching another observation digital twin to the observation window which is the time after the meal / correction insulin is consumed.

[0085] Identify optimal steady-state digital twin for each meal / correction dose: The steadystate digital twin (DTSS-CR) can be selected based on how well the steady-state CGM for a twin matches the real -world CGM in the one hour before the start of the meal. Each steady state digital twin is represented by a given set of insulin sensitivity factors (sf1, sf2, and sf3) as well as total daily insulin requirement (TDIR) and a weight (w). In this way, it is possible write each DTss for a given meal (m) is a function of the insulin sensitivity factors, TDIR, and weight. The following equations of DT as a function of (m) are similar to metabolic model 200 (FIG.2), but they use a different notation to indicate that it would get a different digital twin for each meal (m), whereas metabolic model 200 and related passages above represents a different digital twin for each exercise event.Equation 13DTss(m) = f(sfl,sf2,sf3, TDIR,w)

[0086] Identify optimal observation digital twin for CR and CR recommended changes: Once the optimal DTss is selected, the digital twin for each observation window will be selected (DTobs). The real-world CR and CF can be used to determine the optimal DTobs across all twins. The DTobs can be selected based on minimizing the RMSE between the real-world glucose and glucose forecasted by each digital twin. The Agand tAg parameters are used to identify the DTobs. The Agand tAg parameters from will be optimally selected to identify the DTobs by selecting the combination of Agand tAg parameters that most closely match the real-world glucose data. Agcan vary from zero to 1.4 while tAg can vary from 20 to 80 minutes. In this way, the DTobs will be a function of all of the parameters of the DTSSand also the parameters of Agand tAg.Equation 14DTobs(m) = f(Ag, tAg, sf1, sf2, sf3, TDIR, W)Docket No. 503478.70111:(Tech ID 3305)

[0087] Identification of CR recommendation: If carbohydrates were included in the bolus calculator for a short-acting insulin dose, and if the observation is determined to be eligible per the above criteria, a recommended CR can be calculated by presenting the DTobs with the carbohydrates for that meal event and also presenting the meal insulin as determined using candidates for new CRs (CRcand) which can be a factor of 0.8, 0.9, 1.0, 1.1, and 1.2 of the original CR. Importantly, no correction insulin can be given to the twin when determining the optimal CR. Correction insulin is additional insulin that the person may take to bring their glucose down to a target glucose level. This insulin is not taken to handle the carbohydrates in the meal. During the observation window, the baseline glucose can be determined (Gbaseline = CGM at the start of the meal for the twin) along with the minimum glucose during the observation window (Gmin). Gmin should only be considered if it is at a minimum time (e.g., 20 minutes) after the baseline CGM. An optimal CR is a value that can return the person’s glucose to Gbaseiine. The score should indicate if the meal insulin was able to return the glucose to the baseline. The score could be designed to be higher if the glucose does not return to baseline. Since low glucose is more dangerous than high glucose, if the glucose drops below baseline in the observation window, then the score can be penalized by a factor of two compared with if the glucose returns to a value above the baseline. The score for a CR for observation window (w) is in Equation 15 below.Equation 15^min ^baseline' ^min — ^baseline Score(CR, w)2 X (G baseline ^min < ^baseline

[0088] The optimal CR (CRopt) for that observation window could be the one that corresponds with the minimum score across all CR candidates.Equation 16CRopt(w) = CR -> min(Score(:,w)

[0089] Identification of CF recommendation: If correction insulin was included in the bolus calculator for a short-acting insulin dose, and if the observation is determined to be eligible per the above criteria, a recommended CF can be calculated by presenting the DTobs with the correction insulin for that short-acting insulin dose. Importantly, no carbohydrates and no meal insulin can be given when determining the CF recommendation for the twin. During theDocket No. 503478.70111:(Tech ID 3305)observation window, the baseline glucose can be the glucose at the point when the correction insulin dose was taken (Gbaseiine = CGM at the start of the correction insulin dose for the twin). The minimum glucose (Gmin) can be the minimum CGM during from the time of the shortacting insulin dose + 20 minutes to the duration of the observation window. An optimal CF can return the person to their target glucose (Gtarget) for a given recommendation time frame without bringing them lower than their target. The score can indicate if the correction insulin was able to bring the twin’s glucose to the target glucose. The score can be higher if the glucose for the twin does not reach within ± 10 mg / dL of the Gtarget. Since dropping below the target is more dangerous than bringing them above their target, there can be a higher penalty if the glucose drops more than 10 mg / dL below their target glucose. The score for a CF observation window (w) is given in Equation 17 below.Equation 176 min G targetP > '■’min ’’target '’’target Score(CF, w) = < 2 x (Gitarget ^min) _ 1 f)ov’ "min '■’target” target^target 10 < Gm(n— ^target 1” 10

[0090] The score can range between zero (close to target) to nine (maximally distant from target with target of 40). A typical value for a score could be with a target of 110 and a minimum glucose in the observation window of 200 mg / dL, so the score would be (200-110) / l 10 = 0.818.Determining the final recommended CR and CF based on optimization of net benefit:

[0091] A given CR and CF across an observation window can generate a digital twin glucose response (Gdt(CR, CF, w)) that is a function of that CR and CF. This digital twin glucose response is calculated by using the digital twin to obtain the response to the meal (m), meal insulin (im), and correction insulin (ic), (Gdt(CR, CF, m, im, ic, w)).

[0092] The benefit for a given CR and CF for a given time window, b(CR, CF,w), is calculated by observing the CGM within the observation window and then calculating the score for that window.Docket No. 503478.70111:(Tech ID 3305)Equation 18b(CR, CF,w) = — (score(CR, w) + score(CF, w))

[0093] There is also a benefit of the current CR and CF for that window.Equation 19b CRcurrent, CFcurrent,w') (score(CR current’ w) + score(CFCU7•rent’ w))

[0094] The benefit of selecting a new CR / CF compared with the current CR / CF is the difference between the benefit for the new CR / CF and the original CR / CF. This term is called the delta benefit (A ).Equation 20b(CR, CF,w) = b(CR, CF,w) ~ b CRcurrent, CFcurrent,w)

[0095] The delta benefit is qualified by one minus the normalized error of the glucose predicted by the digital twin (GDT) using the current CR and CF in predicting the actual glucose (Gactuai). The error of the digital twin can be normalized to a value between zero (highest error) and one (no error). Since the maximum error that the digital twin can have is 360 mg / dL (i.e.400-40), the error can be normalized by 360. The error (err) can be determined as follows.Equation 21G actual I err(Gdt, Gactuai,w, C) - - — -

[0096] The error for the current CR and CF is averaged across all time points in the observation window.Equation 22^N=obs \Gdt(t) ~ Gactual(t) err(Gdt, Gactual,w') = — - -

[0097] The expected benefit (eb) is then the delta benefit multiplied by one minus the normalized error as follows.Docket No. 503478.70111:(Tech ID 3305)Equation 23eb(CR, CF) = (1 - err(DTobs)) × b(CR, CF)

[0098] With reference to FIG. 8, consider the following example. A person’s target is 110 mg / dL. They take an insulin bolus that comprises 50% carbohydrate and 50% correction insulin. Their starting glucose when they take the bolus is 220 mg / dL.

[0099] First consider a candidate CR. A candidate CR brings the DT’s glucose to a minimum of 240 mg / dL during the observation window while the DT’s glucose for the DT’s current CR yields a minimum glucose of 260 mg / dL. The score of the candidate CR is (240-220) / 220 = 0.09, and the score for the current CR is (260-110) / 220 = 0.18. The lower score means that the new CR is better than the current CR.

[0100] Next, a candidate CF is considered. The original CF brings the DT’s glucose to a minimum of 140 while the current CF brings the DT’s glucose to a minimum of 180. Therefore, the score for the candidate CF is (140-110) / l 10=0.27 and the score for the current CF is (180— 110) / l 10 = 0.64. The lower score for the candidate CF indicates that it is better than the current CF. The benefit of the current CF and CR is -(0.18+0.63) = -0.81 while the benefit of the candidate CF and CR is -(0.09+0.27) = -0.36. Notice that there is more benefit for the candidate CF and CR. The delta benefit of the candidate CR and CR is (-0.36 - (-081)) = 0.45.

[0101] Now consider that the error is high for this digital twin for this observation, meaning that there may not be much trust (e.g., err = 300 mg / dL off for each time point). The error will be err = 0.833 (e.g., 300 / 360). Then the qualifier for modifying the benefit is 1 - err = 0.167. The expected delta benefit for this CR candidate given both the delta benefit and the error of the DT is the 0.167 * 0.45 = 0.075.

[0102] To see the affect that the error on the digital twin has on the delta benefit, consider the exact same conditions above, but now consider that the error is very small (e.g., err = 20 mg / dL at each time point during the observation window). In this case, the scores and the delta benefit are all identical with the example above. But the 1 - err term is now 0.94. Therefore, the expected benefit is 0.429. In this way, it is possible to use the (1 - error) term to weight the result from the second observation by five to six times as much as in the first example, since the error is so much smaller.

[0103] As another example, consider the case where the CR and CF candidates make things worse compared with the current CR and CF settings (i.e., the candidate CR does not bring theDocket No. 503478.70111:(Tech ID 3305)glucose close to the starting glucose and the candidate CF does not bring the glucose close to the target glucose). This will cause a negative benefit. And this negative benefit will be considered worse if there is less error in the DT. Consider the same starting glucose and target as in the example above. But now consider that the original CR returns the glucose to 240 mg / dL in the observation window while the candidate CR returns the glucose to 260 mg / dL. And consider that the original CF bring the glucose to 140 mg / dL while the new CF brings the glucose to 180 mg / dL. Now also consider that there is a low error observation and then a high error observation. The results of the calculations of expected benefit are shown in FIG. 8 for Observations 3 (high error DT) and 4 (low error DT). Notice that the negative benefit is larger for the case where there is lower error for the DT, since there is more trust than the higher error DT observation.

[0104] For each observation window, it is possible to create a table of expected benefit values for each of the various CR and CF combinations and multiplied by the error for the correction digital twin and the error for the carb ratio digital twin. Across each of the observation windows, it is possible to average the table entries point-wise. The combination with the largest eb can be selected.Table 2: List of expected benefit (eb) that a person using the decision support system would expect for different CR and CF combinations that are a fraction of the current CR and CF.CF\CR0.9 1.0 1.1 1.20.8 eb(0.8,0.8) eb(0.8,0.9) eb(0.8,1.0) eb(0.8,l.1) eb(0.8,1.2) 0.9 eb(0.9,0.8) eb(0.9,0.9) eb(0.9,1.0) eb(0.9,l.l) eb(0.9,1.2) 1.0 eb(1.0,0.8) eb(1.0,0.9) eb(l.0,1.0) eb(l.0,1.1) eb(1.0,1.2) 1.1 eb(l.1,0.8) eb(l.1,0.9) eb(l.1,1.0) eb(l.1,1.1) eb(l.1,1.2) 1.2 _ eb(1.2,0,8) eb(1.2,0,9) eb(1.2, L0) eb(l.2,1,1) eb(1.2, L2)

[0105] Alternatively, the selection of the CR and CF that conferred the maximum benefit can be done separately. For selecting CR, the CF can be kept constant at the current setting (i.e., CF adjustment factor = 1.0) and variations are considered only for adjustments to insulin-to-carbohydrate ratio, resulting in a row vector of expected benefit values selected from Table 2 when CF = 1.0 and the CR factor varies from 0.8 to 1.2. Similarly, for selecting CF, CR can be kept constant at the current setting (i.e., CR adjustment factor = 1.0) and variations are considered only for adjustments to the set correction factor, resulting in a column vector of expected benefit values selected from Table 2 when CR = 1.0 and the CF adjustment factor varies from 0.8 to 1.2. In both cases, the combination that yields the highest benefit is selected.Docket No. 503478.70111:(Tech ID 3305)

[0106] Explainability of the recommendation is included by using the Ambulatory Glucose Profile (AGP) metrics, which can be calculated from the digital twin glucose trace (Gat) and compared with the real world glucose trace (Gactuai) for each CR and CF selected. The metrics of interest can include the LBGI, HBGI, TIR, TBR, TAR, and mean glucose across the observation window.Table 3: Metrics used to describe benefit of use of a new CR, CF, or long-acting insulin dose.Metric _Glucose-index HBGlDt-HBGI Actual LB GI Dt— LB GI Actual Time in range (TIR: 70-180 mg / dL) TIRDT- TIRActualTime below range (TBR: < 70 mg / dL) TBRDT-TBRActualTime above range (TAR: > 180 m / dL) T ARDT-T ARActualMean glucose MeanCGMoT-MeanCGMActuaiLong-acting insulin modification recommendations

[0107] The long-acting insulin dose is given once or twice each day. An app that implements this DSS (e.g., DailyDose) can provide recommendations on modifying the long-acting insulin dose for participants also using digital twin approaches.

[0108] The long-acting observation window can be at least four hours after a carbohydrate intake or a short-acting insulin dose. The observation window for long-acting insulin can be at least one hour in duration.

[0109] Selection of optimal digital twin and selection of recommendation: The long-acting digital twin for each observation window can be selected from one of the 99 digital twins that are in the OHSU metabolic simulator.

[0110] Observation windows can be considered throughout all time periods in a 24-hour day. A digital twin can be identified for every eligible long-acting insulin observation window. A long-acting insulin observation window can be considered eligible for a possible recommended change if the following criteria are met: (1) The observation window starts at least four hours after the most recent carbohydrate intake and at least four hours after the most recent shortacting insulin dose. (2) The observation window is at least one hour in duration. (3) The observation window is no longer than 12 hours in duration. (4) There is long-acting insulin on board at every time point in the observation window. The long-acting insulin on board isDocket No. 503478.70111:(Tech ID 3305)approximated using the following methods depending on if the person using the system is on once-per-day or twice-per-day long acting insulin dosing.

[0111] Once-per-day long-acting insulin on board can be approximated by presuming that the long-acting insulin dose is delivered equivalently every five minutes across a 24-hour period in the amount of long_acting_insulin_dose / 288.

[0112] Twice-per-day long-acting insulin on board can be approximated by presuming that the long-acting insulin dose is delivered equivalently over a 12-hour period in the amount of long_acting_insulin_dose / 144.

[0113] Determination of optimal long-acting digital twin: The optimal long-acting digital twin can be identified by matching a digital twin to the long-acting observation window (DTobs) using minimal RMSE of the glucose produced by each digital twin compared with the real-world glucose. Each long-acting digital twin is represented by a given set of insulin sensitivity factors (sf1, sf2, and sf3) as well as TDIR and a weight (W). In this way, it is possible write each DTiong acting for a giving observation window (w).Equation 24DTlong acting(w) = f(sf1,sf2,sf3, TDIR, W)

[0114] Identification of long-acting dose recommendation: If the long-acting observation window is considered to be eligible, the digital twin can be identified and then presented with different candidate long-acting insulin doses that can be a factor of 0.8, 0.9, 1.0, 1.1, and 1.2 of the original long-acting insulin dose. The optimal long-acting insulin dose for that observation window can be the one that achieves the best score that brings the digital twin’s glucose to a target value (Gtarget) at the end of the observation window (Gend). The score for the long-acting insulin observation window (w) is given below:Equation 25= p XI 0. G,. - 10 < G < G„ » 4- 10Docket No. 503478.70111:(Tech ID 3305)

[0115] A lower score is determined by enabling the digital twin’s glucose to more closely reach the target glucose by the end of the observation window. A score of zero means that the basal rate has kept the person’s glucose exactly at the target glucose level.

[0116] The final recommended long-acting dose can be selected based on optimization of net benefit. A given long-acting insulin dose for an observation window can generate a digital twin glucose response (Gdt(dose, w)) that is a function of that dose.

[0117] The benefit for a given long-acting dose for a given time window, b(dose,w), is calculated by observing the CGM within the observation window and then calculating the score for that window.Equation 26b(dose, w) = —score (dose, w)

[0118] There is also a benefit of the current long-acting dose for that window.Equation 27b(dosecurrent,w') = score(dosecurrent,w)

[0119] The benefit of selecting a new long-acting insulin dose compared with their current long-acting insulin dose is the difference between the benefit for the new dose and the original dose. This term is called the delta benefit (Ab).Equation 28Ab(dose,w) = b(dose, w) — b(dosecurrent, w)

[0120] The delta benefit is qualified by the error of the glucose predicted by the digital twin (Gdt) using the current long-acting insulin dose in predicting the actual glucose (Gactuai). The error of the digital twin can be normalized to a value between zero (highest error) and one (no error). Since the maximum error that the digital twin can have is 360 mg / dL (i.e. 400-40), the error can be normalized by 360. The error (err) can be determined as follows.Equation 29I G<it Gactuai|err(G^t, Gact:uai, w, t)Docket No. 503478.70111:(Tech ID 3305)

[0121] The error for the current long-acting insulin dose is averaged across all time points in the observation window.Equation 30yJV=obs \Gdt(t) - Ggctugl G) Iprr(C C Gactuaj, w) _ 360 _

[0122] The expected benefit (eb) is therefore the delta benefit multiplied by the normalized error for that observation window as follows.Equation 31eb(dose) = (1 - err(DTobs, Gactuai,w^ X Ab(dose.w)

[0123] For each observation window, it is possible to calculate which long-acting insulin dose yields the maximum expected benefit, and this can be the long-acting insulin dose recommended for use by the person using the app. Across each of the observation windows, it is possible to average the table entries point-wise. The combination with the largest eb can be selected.

[0124] Explainability of the recommendation is included by using the AGP metrics described above in Table 3.Example System and Device Implementations

[0125] FIG. 9 shows a cloud-connected diabetes management system 900, which supports insulin delivery optimization by processing CGM, insulin, and exercise data to enhance realtime glucose control and treatment interventions. System 900 integrates with both automated insulin delivery (AID) systems and connected insulin pens (both of which may be referred to generally as connected insulin delivery systems), adjusting insulin dosing dynamically to optimize glucose outcomes. By leveraging real-time physiological data and digital twin-based metabolic models, system 900 improves glucoregulatory hormone dosing by enabling automated updates and optimization of the glucoregulatory hormone dosage parameters, making it a comprehensive, multi-functional diabetes management platform., making it a comprehensive, multi-functional diabetes management platform. More generally, connected insulin delivery systems are types of connected glucoregulatory hormone delivery systems for delivery of one or more of insulin, glucagon, and pramlintide.Docket No. 503478.70111:(Tech ID 3305)

[0126] In AID system embodiments, system 900 automatically adjusts insulin dosing by transmitting optimized insulin delivery parameters to automated pumps, ensuring real-time closed-loop control. In connected insulin pen (or smart pen) embodiments, system 900 provides insulin dosing adjustments based on predicted glucose trends, allowing users to administer the optimal insulin dose using the optimized settings of the insulin dosing system.

[0127] Specifically, system 900 includes a medical device 902 — which may be a CGM 904, a connected insulin pen 906, or an automated insulin pump — a user’s software application 908 (e.g., a smartphone, smartwatch, or other smart device app), and a cloud-based software application 910 running on associated computing devices. Medical device 902 communicates real-time glucose and insulin delivery data via a personal area network (PAN) connection 912 (e.g., Bluetooth) to software application 908.

[0128] Software application 908 generates a user interface 914 that provides real-time feedback and insulin adjustment insights derived from system 900. FIG. 9 depicts core algorithms 916, including exercise-aware decision support (exDSS) 918, periodic insulin adjustment DSS 920, and meal detection algorithms 922 that help optimize insulin delivery timing and carbohydrate intake management. For completeness, software application 908 also includes lower-layer OS components such as network stack 924, which manage connectivity and data synchronization.

[0129] Software application 910 receives data from application 908 via a secure internet connection 926, processes it using digital twin-based predictive models, and stores it in data storage 928 for advanced analytics. The processed data is then used to generate data visualizations 930 on user interface 914, allowing users and healthcare providers to track trends, optimize dosing strategies, and improve long-term glucose management.

[0130] FIG. 10 is a block diagram illustrating components 1000, according to some example embodiments, able to read instructions from a machine-readable or computer-readable medium (e.g., a non-transitory machine-readable storage medium), and perform any one or more of the methods discussed herein (e.g., FIG. 11). For example, hardware resources 1002 may be embodied in a smartwatch, server, tablet computer, or patient-connected device that provides the ability to measure a physiological signal, provide some analysis of that signal, transmit information about that signal, and / or support a user interface to provide information about that signal. This includes an equivalent functional combination, for example a watch that canDocket No. 503478.70111:(Tech ID 3305)measure a physiological signal and transmit the data to a computer (including a smartphone), where the computer provides analysis and user interface functions.

[0131] Specifically, FIG. 10 shows a diagrammatic representation of hardware resources 1002 including one or more processors 1004 (or processor cores), one or more memory / storage devices 1006, and one or more communication resources 1008, each of which may be communicatively coupled via a bus 1010.

[0132] Processors 1004 (e.g., a central processing unit (CPU), a reduced instruction set computing (RISC) processor, a complex instruction set computing (CISC) processor, a graphics processing unit (GPU), a digital signal processor (DSP) such as a baseband processor, an application specific integrated circuit (ASIC), another processor, or any suitable combination thereof) may include, for example, a processor 1012 and a processor 1014.

[0133] Memory / storage devices 1006 may include main memory, disk storage, or any suitable combination thereof. Memory / storage devices 1006 may include, but are not limited to any type of volatile or non-volatile memory such as dynamic random access memory (DRAM), static random-access memory (SRAM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), Flash memory, solid-state storage, etc.

[0134] Communication resources 1008 may include interconnection or network interface components or other suitable devices to communicate with one or more peripheral devices 1016 or one or more databases 1018 via a network 1020. For example, communication resources 1008 may include wired communication components (e.g., for coupling via a Universal Serial Bus (USB)), cellular communication components, NFC components, Bluetooth® components (e.g., Bluetooth® Low Energy), Wi-Fi® components, and other communication components.

[0135] Instructions 1022 may comprise software, a program, an application, an applet, an app, or other executable code for causing at least any of processors 1004 to perform any one or more of the methods discussed herein (e.g., process 1100, FIG. 11). In some embodiments, instructions 1022 include entail selecting a diabetes treatment intervention for a person using MDI insulin therapy for T1D or T2D, which are stored in memory / storage devices 1006.

[0136] Instructions 1022 may reside, completely or partially, within at least one of processors 1004 (e.g., within the processor’s cache memory), memory / storage devices 1006, or any suitable combination thereof. Furthermore, any portion of instructions 1022 may be transferred to hardware resources 1002 from any combination of the peripheral devices 1016 or theDocket No. 503478.70111:(Tech ID 3305)databases 1018. Accordingly, the memory of processors 1004, memory / storage devices 1006, peripheral devices 1016, and databases 1018 are examples of computer-readable and machine-readable media.

[0137] FIG. 11 shows a process 1100 of selecting a diabetes treatment intervention for a person using MDI insulin therapy for T1D or T2D. In this example, process 1100 is performed by computing device 102 (FIG. 1) configured with a DSS app 104 that uses the optimized settings determined by the digital twin. Although this example is specific to insulin, it may be applied to other glucoregulatory hormone treatments.

[0138] In block 1102, process 1100 receives glucose data generated by a CGM coupled to the person. In block 1104, process 1100 receives insulin data generated by an insulin pen configured to deliver insulin to the person. In block 1106, process 1100 receives heart rate data generated by a heart rate monitor coupled to the person. In block 1108, process 1100 receives carbohydrate consumption data representing carbohydrates consumed by the person. In block 1110, process 1100 in response to a scheduled event or an on demand request initiated by the person via a user interface of the DSS application, determines from a set of digital twins, based on the glucose, insulin, heart rate, and carbohydrate consumption data, a selected digital twin that acts as a personalized metabolic model of glucose dynamics of the person. In block 1112, process 1100 initializes states of the selected digital twin to establish an initialized digital twin. In block 1114, process 1100 simulates a set of available diabetes treatment interventions as inputs for the initialized digital twin to compare each calculated diabetes treatment benefit associated with the set of available diabetes treatment interventions. In block 1116, process 1100 presents to the person, via the user interface, the diabetes treatment intervention yielding an extremum calculated diabetes treatment benefit or other desired target or threshold (range) determined from among the each calculated diabetes treatment benefit.

[0139] In other embodiments, computing device 102 receives data and transmits it to a remote server for further processing. In that embodiment, the remove server determines a digital twin, initializes its states, and so forth. In still other embodiments, different processing steps are performed by each of computing device 102 and remote server, depending on the particular configuration.

[0140] Skilled persons will appreciate that many changes may be made to the details of the above-described embodiments without departing from the underlying principles of theDocket No. 503478.70111:(Tech ID 3305)invention. The scope of the present invention should, therefore, be determined only by claimed inventions and equivalents thereof.

Claims

Docket No. 503478.70111:(Tech ID 3305)CLAIMSWhat is claimed is:

1. A diabetes management system for optimizing glucoregulatory hormone delivery, the system comprising:a connected glucoregulatory hormone delivery system; anda computing device comprising a processor and a non-transitory computer-readable medium storing instructions that, when executed by the processor, configure the diabetes management system to:receive glucose management data from the connected glucoregulatory hormone delivery system or associated sensors;select a digital twin metabolic model based on the received data and initialize its states;simulate diabetes treatment interventions using the digital twin metabolic model to evaluate predicted glucose dynamics;determine a glucoregulatory hormone dosage parameter based on an intervention yielding an extremum calculated diabetes treatment benefit; andtransmit the glucoregulatory hormone dosage parameter to the connected glucoregulatory hormone delivery system for glucoregulatory hormone delivery.

2. The system of claim 1, in which the glucose management data includes glucose and insulin data.

3. The system of claim 2, in which the diabetes management system continuously updates the digital twin metabolic model using real-time glucose and insulin data to provide predictive adjustments to the glucoregulatory hormone delivery.

4. The system of claim 1, in which the glucose management data includes heart rate data.

5. The system of claim 4, further comprising a wearable fitness device that provides the heart rate data, in which the diabetes management system adjusts insulin delivery based on predicted glucose changes due to physical activity.

6. The system of claim 1, in which the glucose management data includes carbohydrate consumption data.Docket No. 503478.70111:(Tech ID 3305)7. The system of claim 1, in which the connected glucoregulatory hormone delivery system comprises a glucoregulatory hormone pump.

8. The system of claim 7, in which the glucoregulatory hormone pump autonomously adjusts the glucoregulatory hormone delivery in response to the glucoregulatory hormone dosage parameter.

9. The system of claim 1, in which the connected glucoregulatory hormone delivery system comprises a connected glucoregulatory hormone pen system.

10. The system of claim 9, in which the connected glucoregulatory hormone pen system displays the glucoregulatory hormone dosage parameter on a user interface for manual confirmation before injection.

11. The system of claim 1, in which the associated sensors include a continuous glucose monitor (CGM).

12. The system of claim 1, in which the associated sensors include an insulin relay PDM or smart device.

13. The system of claim 1, in which the diabetes management system is further configured to select the glucoregulatory hormone dosage parameter that minimizes hypoglycemia or hyperglycemia risk.

14. The system of claim 1, in which the diabetes management system is further configured to determine the glucoregulatory hormone dosage parameter before, during, or after exercise based on heart rate data and carbohydrate intake.

15. The system of claim 1, in which the glucoregulatory hormone dosage parameter is made periodically for long-term glucose control.

16. A method for optimizing glucoregulatory hormone delivery to a person, performed by a diabetes management system comprising a computing device configured with a diabetes treatment application, the method comprising:receiving glucose management data from a connected glucoregulatory hormone delivery system or associated sensors;Docket No. 503478.70111:(Tech ID 3305)determining, in response to a scheduled event or an on-demand request, from a set of digital twins, a selected digital twin that acts as a personalized metabolic model of glucose dynamics of the person based on the received glucose management data;initializing states of the selected digital twin to establish an initialized digital twin; simulating diabetes treatment interventions using the initialized digital twin to calculate diabetes treatment benefit and evaluate predicted glucose dynamics;determine a glucoregulatory hormone dosage parameter based on a simulated intervention yielding an extremum calculated diabetes treatment benefit; andtransmit the glucoregulatory hormone dosage parameter to the connected glucoregulatory hormone delivery system for delivery.

17. The method of claim 16, in which the glucose management data includes glucose levels from a continuous glucose monitor (CGM) coupled to the person.

18. The method of claim 16, in which the glucose management data includes insulin delivery data from the connected glucoregulatory hormone delivery system.

19. The method of claim 16, in which the glucose management data includes heart rate data from a heart rate monitor coupled to the person.

20. The method of claim 16, in which the glucose management data includes carbohydrate consumption data representing carbohydrates consumed by the person.

21. The method of claim 16, in which the personalized metabolic model of glucose dynamics includes a set of ordinary differential equations (ODEs).

22. The method of claim 16, in which the connected glucoregulatory hormone delivery system is at least one of an automated insulin delivery (AID) system including an insulin pump or a connected insulin pen system.

23. The method of claim 16, in which the connected glucoregulatory hormone delivery system is at least one of an automated hormone delivery (AHD) system including a glucoregulatory hormone pump or a connected glucoregulatory hormone pen system for one or more of insulin, glucagon, or pramlintide administration.Docket No. 503478.70111:(Tech ID 3305)24. The method of claim 16, in which the calculated diabetes treatment benefit is a sum of LBGI and HBGI.

25. The method of claim 24, in which the sum of LBGI and HBGI is a minimum among each calculated diabetes treatment benefit corresponding to the diabetes treatment interventions.

26. The method of claim 16, in which the calculated diabetes treatment benefit comprises one or more of the Ambulatory Glucose Profile metrics.

27. The method of claim 16, in which the calculated diabetes treatment benefit corresponds to a carbohydrate-to-insulin (CR) adjustment score that is based on a difference between minimum and baseline glucose.

28. The method of claim 16, in which the calculated diabetes treatment benefit corresponds to a correction factor (CF) adjustment score that is based on a difference between minimum and baseline glucose.

29. The method of claim 16, in which the calculated diabetes treatment benefit corresponds to a combined CR and CF adjustment score that are based on a difference between minimum and baseline glucose.

30. The method of claim 16, in which the diabetes treatment interventions include a change in exercise duration.

31. The method of claim 16, in which the diabetes treatment interventions include a change in exercise intensity.

32. The method of claim 16, in which the diabetes treatment interventions include an amount of carbohydrates to eat.

33. The method of claim 16, in which the diabetes treatment interventions include a time to wait to exercise.

34. The method of claim 16, in which the diabetes treatment interventions include a change in CR for calculating short-acting meal insulin.Docket No. 503478.70111:(Tech ID 3305)35. The method of claim 16, in which the diabetes treatment interventions include a change in basal insulin infusion rate.

36. The method of claim 16, in which the diabetes treatment interventions include a change in correction factor for calculating short-acting insulin when glucose is too high.

37. The method of claim 16, in which the diabetes treatment interventions include a change in the long-acting insulin.

38. The method of claim 16, in which the diabetes treatment interventions include change of behavior before, during, or after exercise by following one or more of recommendation including consume a carbohydrate, reduce or increase a short-acting insulin dose, delay exercise, or do a different form of exercise.

39. The method of claim 16, in which the diabetes treatment interventions include a periodic recommendation for adjusting one or more of CR, CF, and long-acting insulin.

Images