System and method to predict glucose response

Abstract

The present disclosure relates to a system and method to predict the post- prandial glucose response in the individuals type 2 diabetes mellitus (T2DM) in India by taking sample of patients with T2DM from various locations across India and fitting them with continuous glucose monitors (CGM) and calculating the postprandial glucose response based on the different meal intakes by the patients. The present disclosure further relates to the utilization of k-means clustering models to classify food types based on nutritional information and classification of patients. The present disclosure also relates to utilizing XGBoost to predict postprandial blood glucose responses based on patient phenotypes and food categories and providing personalized recommendations taking into account meal type, preferences, and regional influences which would allow patients with T2DM to eat and drink foods to maintain their blood sugar within prescribed limits and prevent disease progression.

Description

[0001] SYSTEM AND METHOD TO PREDICT GLUCOSE RESPONSE

[0002] TECHNICAL FIELD:

[0003] The present disclosure relates to the field of healthcare and medical diagnostics. More particularly, the present disclosure relates to a system and method for processing phenotype and nutritional information to generate personalized recommendations for reducing blood sugar impact of food item on individuals with type 2 diabetes mellitus (T2DM).

[0004] BACKGROUND:

[0005] Background description includes information that may be useful in understanding the present disclosure. It is not an admission that any of the information provided herein is prior art or relevant to the presently claimed disclosure, or that any publication specifically or implicitly referenced is prior art.

[0006] Type 2 diabetes mellitus (T2DM) is a serious health concern worldwide. India ranks second globally in the number of individuals affected by T2DM, with an estimated 74.9 million cases in 2021, projected to rise to 124.9 million by 2045. Approximately one in seven diabetic adults worldwide lives in India, with one in every three households having a diabetic member. A population - based study conducted between 2012 and 2014 estimated a 7.5% T2DM prevalence. The Indian Council of Medical Research conducted a study from 2008 to 2015, revealing varying T2DM prevalence among states and higher rates among low socioeconomic status groups in urban areas of developed states. Over time, the prevalence of T2DM among Indian adults aged 20 years and older increased from 5.5% in 1990 to 7.7% in 2016. Recent reports from the International Diabetes Federation indicate a T2DM prevalence of 9.3% in 2018 and 9.6% in 2021, respectively, with projections indicating a rise to 10.4% by 2030 in Indians. Certain factors evidence that the Indian population is more prone to T2DM when compared to the population from other parts of the world, such as:

[0007] • Onset of T2DM at a younger age.

[0008] • Lower body -mass index (BMI) when compared to the population from other parts of the world.

[0009] • Higher levels of insulin resistance (and for longer periods of time).

[0010] • Premature beta-cell failure.

[0011] • Higher chances of developing the fatal complications of T2DM.

[0012] These elements stem from a mix of lifestyle choices, epigenetics, and prenatal influences within the Indian population. Due to the varied effect of these factors on the Indian population and the variations in Indian food type compared to other parts of the world, there are differences in how blood sugar levels respond to a particular food type in different individuals.

[0013] Present medical recommendations for managing disease require the patients to meticulously monitor their calorie intake from food and implement significant alterations to both their diet and lifestyle. Most individuals with T2DM encounter challenges in implementing and maintaining such modifications, with only a minute proportion achieving sustained adherence. Moreover, the uniformity of advice provided fails to consider the diverse aspects of the condition, thereby oversimplifying its management.

[0014] As a result, effective management of T2DM remains challenging, and its burden both in terms of prevalence and its impact on individuals and healthcare systems continues to grow steadily. Because the condition is typically chronic in nature, management of the disease requires measuring and controlling the body’s blood sugar response to food (referred to as postprandial glucose response). Post-prandial glucose response is typically measured after consumption of the food, due to which the patients with T2DM face a challenge in ascertaining and accurately predicting it in advance. Moreover, ascertaining the blood glucose response to a particular food with accuracy at an individual level is difficult as the phenotypic differences greatly contribute to it due to which there are marked inter-individual differences in post-prandial glucose response.

[0015] Models to predict post-prandial glucose response are limited, and available models suffer from inaccuracy and heterogeneity which typically rely on a linear regression model to predict the response in T2DM patients as these models are designed to predict the identical post-prandial glucose response for the same food item from all individuals with T2DM without characterization and identification of the factors associated with the variability in post-prandial glucose response among individuals with T2DM in India. One of the fundamental limitations of existing methods is their inability to differentiate between different types of food or account for the diverse variability in responses exhibited by T2DM patients to the same food. The linear regression approach used in these models lacks the sophistication needed to capture the non-linear variability in food types and individual responses. Hence, they fail to capture the inter-individual phenotypes due to which the output of these models is generic, and not very accurate. Therefore, these models are less reliable for individuals having significant variation in the blood glucose response due to their phenotypic differences. Furthermore, these models are often trained by datasets from non-Indians, whose genetic make-up varies significantly from that of Indians, adding to the inaccuracy in predictions.

[0016] There is, therefore, a need to provide a method to accurately predict the postprandial glucose response for a given Indian T2DM patient, which takes into account the variation in the different types of Indian foods and the interindividual variation in the post prandial glucose response in the Indian T2DM patients so that recommendations can be personalized based on nutritional status, lifestyle, and metabolic goals of the patients.

[0017] SUMMARY:

[0018] The present invention provides a system and method for predicting postprandial glucose response (PPGR) in individuals, particularly those with type 2 diabetes, that includes multiple artificial intelligence (Al) and machine learning models. The system comprises a memory unit configured to store phenotype and nutritional information, and a processing unit operatively coupled with the memory unit. The processing unit executes multiple Al models, including an Al Grouping Model for classifying phenotype information into clusters, and a Nutrition Al Model for classifying nutritional information into one from group of distinct clusters.

[0019] Based on these classifications, a supervised Al model predicts the PPGR of an individual with a defined granularity, represented on a graded scale. The predicted PPGR output is then used to generate personalized recommendations aims at at reducing the blood sugar impact of food items on the individual. The recommendations may include dietary modifications, food substitutions, or lifestyle actions such as exercise timing, optimized according to individual preferences and regional meal habits.

[0020] The invention thus provides a robust, data-driven approach for managing blood glucose responses through personalized dietary and lifestyle guidance. By integrating phenotype information, nutritional analysis, and supervised learning models, the system enhances the accuracy of PPGR prediction and supports proactive diabetes management tailored to individual needs.

[0021] One of the aspects of the present disclosure relates to taking a sample of patients with T2DM from various locations across India and fitting them with continuous glucose monitors (CGM) providing specific protocols for meal consumption and activity levels, thereby collecting the data by logging dietary intake.

[0022] Another aspect of the present disclosure relates to the utilization of k-means clustering models to classify food types based on nutritional information.

[0023] Another aspect of the present disclosure relates to train the supervised learning models, such as XGBoost to predict postprandial blood glucose responses based on individual patient phenotypes and food categories.

[0024] Yet another aspect of the present disclosure relates to providing personalized recommendations to individuals with T2DM to reduce blood sugar impact, based on meal type, preferences, and regional influences which would allow patients with T2DM to eat and drink foods to maintain their blood sugar within prescribed limits and prevent disease progression.

[0025] BRIEF DESCRIPTION OF THE ACCOMPANYING DRAWINGS

[0026] Figure.1 illustrates a flowchart depicting a method for prediction of postprandial glucose response (PPGR) in an individual with type 2 diabetes, according to the disclosure of the present invention.

[0027] Figure.2 illustrates a system for predicting post-prandial glucose response (PPGR) in an individual with type 2 diabetes, according to the disclosure of the present invention.

[0028] Figure.3 illustrates a model design for collecting and monitoring individual health parameters, according to the disclosure of the present invention.

[0029] It should be appreciated by those skilled in art that any diagrams herein represent conceptual views of illustrative systems embodying the principles of the present subject matter. DETAILED DESCRIPTION:

[0030] In the present document, the word "exemplary" is used herein to mean "serving as an example, instance, or illustration." Any embodiment or implementation of the present subject matter described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments.

[0031] The specific embodiment thereof has been shown by way of example in the drawings and will be described in detail below. It should be understood, however that it is not intended to limit the disclosure to the particular forms disclosed, but on the contrary, the disclosure is to cover all modifications, equivalents, and alternative falling within the spirit and the scope of the disclosure.

[0032] In the following detailed description of the embodiments of the disclosure, reference is made to specific embodiments in which the disclosure may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the disclosure, and it is to be understood that other embodiments may be utilized and that changes may be made without departing from the scope of the present disclosure. The following description is, therefore, not to be taken in a limiting sense.

[0033] A system for predicting post-prandial glucose response (PPGR) in an individual with type 2 diabetes comprises a processing unit coupled with a memory unit configured to store a predefined set of data. The processing unit is configured to retrieve the predefined data along with nutrition information of at least one food item and phenotype information of at least one individual. The phenotype and nutrition information are processed separately using multiple Artificial Intelligence (Al) models, wherein the phenotype information is classified into one of a plurality of clusters using an Al grouping model, and the nutritional information is classified into distinct clusters using a nutrition Al model. The combined phenotype and nutrition information is then processed using a supervised Al model to predict PPGR with a defined granularity, represented on a scale of values. The predicted PPGR output is further used to generate personalized recommendations aimed at reducing the blood sugar impact of the food item on the individual.

[0034] A method for predicting post-prandial glucose response (PPGR) in an individual with type 2 diabetes comprises retrieving, by a processing unit from a memory, a predefined set of data including phenotype information of at least one individual and nutrition information of a food item. The phenotype information is processed with an Al grouping model to analyze the phenotype, while the nutrition information is processed with a nutrition Al model to analyze the nutritional data. The Al grouping model classifies the individual into one of a plurality of phenotype clusters, and the nutrition Al model classifies the nutrition information into distinct nutritional clusters. Based on the phenotype cluster and the nutritional cluster, a supervised Al model predicts the PPGR of the individual with a defined granularity on a graded scale. The predicted PPGR is further used to generate personalized recommendations aimed at reducing the blood sugar impact of the food item for the individual.

[0035] Figure 1 illustrates a method (100) for predicting post-prandial glucose response (PPGR) in an individual with type 2 diabetes using artificial intelligence (Al) models. The method (100) is represented by a flowchart for predicting post-prandial glucose response (PPGR) in an individual with type 2 diabetes. The process starts with step (102), where a processing unit retrieves from memory a predefined set of data comprising phenotype information of at least one individual along with nutrition information of a food item. In step (104), the phenotype information is processed using an Al Grouping Model to analyze the individual’s characteristics, while in step (106) the nutrition information is processed using a Nutrition Al Model to analyze nutritional attributes of the food item. In step (108), the Al Grouping Model classifies the individual into one of a plurality of phenotype clusters, and in step (110), the Nutrition Al Model classifies the nutrition information into distinct nutritional clusters. Thereafter, in step (112), a supervised Al model predicts the PPGR of the individual as output with a defined granularity on a graded scale of 1 to 5, based on the classified phenotype cluster or category and nutritional cluster. Finally, in step (114), the supervised Al model further utilizes the classified phenotype and nutritional information to generate personalized recommendations aimed at reducing the blood sugar impact of the food item on the individual. The method then concludes with the generating actionable, individualized or personalized recommendations derived from the PPGR prediction.

[0036] The process, after generating predictions and recommendations, providing a personalized tool for managing post-prandial glucose levels in type 2 diabetic individuals. These recommendations may include actions such as food substitutions e.g., replacing white rice with brown rice, meal sequencing strategies e.g., consuming protein before carbohydrates, or lifestyle adjustments e.g., post-meal walking. The process ends after these predictions and recommendations are provided, creating a tailored approach for managing blood glucose.

[0037] Figure.2 illustrates a system (200) for predicting post-prandial glucose response (PPGR) in an individual with type 2 diabetes, according to the disclosure of the present invention. The system (200) includes a memory unit (202) configured to store predefined datasets, such as phenotype information of an individual and nutritional information of food items. A processing unit (204) is coupled with the memory unit (202). The processing unit (204) interacts with multiple artificial intelligence (Al) models: an Al Grouping Model (206), which processes and classifies the individual’s phenotype information into one of distinct phenotype clusters, and a Nutrition Al Model (208), which processes nutritional information and classifies it into distinct nutritional clusters.

[0038] The individual phenotype (210) and the classification of nutritional information (212) form the processed inputs that are fed into a supervised Al model (214). This supervised Al model combines the phenotype information and nutritional information to generate an output (216) that predicts the PPGR of the individual with a defined granularity. Based on this PPGR prediction output, the system produces personalized recommendations (218), which are aimed at minimizing the blood sugar impact of specific food items on the individual.

[0039] The system is provided with data storage and further retrieval, progressing through Al-driven models and classifications and, prediction using a supervised model. Further, generating actionable dietary recommendations specific to the individual’s health profile.

[0040] In an embodiment, the Nutrition Al Model (208) is implemented as an unsupervised machine learning model, including but not limited to clustering models such as K-Means. In this embodiment, the unsupervised model is configured to classify nutritional information into distinct categories such as high sugar, normal, and high protein.

[0041] In another embodiment, the Al Grouping Model (206) for classifying clinical or phenotype information comprises a supervised machine learning model. The Al Grouping Model is configured to classify an individual into the most relevant phenotype cluster among a plurality of distinct phenotype clusters. In some cases, the Al Grouping Model is further configured to perform clustering based at least on historical blood sugar response data, thereby enhancing the accuracy of classification and prediction.

[0042] In yet another embodiment, the phenotype information associated with an individual comprises at least one of anthropometric measurements, activity level, and disease duration. Such phenotype information may include, but is not limited to, age in years as a diabetic, weight, waist size, hip size, calf size, number of exercise days since last exercise, meal count per week, average sitting time, years as a diabetic, and average sleep duration.

[0043] In another embodiment, the nutritional information comprises attributes such as protein, fat, carbohydrates, fiber, sugar, total calories, and other macronutrient and micronutrient data relevant to glucose response analysis.

[0044] In another embodiment, the supervised Al model (214) configured for predicting post-prandial glucose response comprises a gradient boosting model, including but not limited to an XGBoost model. In some embodiments, the distinct clusters of nutritional information stored in memory are derived from a clinical study quantifying post-prandial glucose responses (PPGR) for Indian-specific meals.

[0045] In another embodiment, the output module (216) is configured to provide the predicted PPGR to the individual on a graphical user interface. In some embodiments, the supervised Al model (214) is further configured to predict the post-prandial glucose response on a graded scale of 1 to 5, wherein the scale represents different levels of blood sugar impact.

[0046] In another embodiment, the system further configured to generate personalized recommendations to reduce the blood sugar impact of a food item on an individual, wherein the recommendations comprise: - identifying a plurality of actions associated with reducing the predicted blood glucose response;

[0047] - ranking the plurality of actions based on factors including at least one of meal type, individual meal preferences, and regional meal preferences; and - generating, based on said ranking, a list of recommended actions with an estimated reduction in blood sugar impact for each action.

[0048] In another embodiments, the actions recommended to reduce blood sugar include substituting one food item with another or performing a physical activity after consumption, wherein the system configured to optimize the prioritize actions with higher reduction impact. Examples of ranked actions include carbohydrate modification (such as substituting brown rice for white rice), meal sequencing (such as delaying carbohydrate consumption until after protein intake), or exercise timing (such as performing post-meal walking). Experimental Data

[0049] A study is conducted to evaluate the PPGR and self-management activities for individuals with diabetes type 2. The design, conduct and analysis of the present study meet International Committee of Medical Journal Editors (ICJME) criteria. Figure 3. illustrates a model design for collecting and monitoring individual health parameters for predicting post-prandial glucose response. Initially, in a clinical setting on Day 1, baseline information is obtained through health surveys, biometric measurements, and laboratory assessments. Following the clinical phase, during an at-home monitoring period spanning Days 1 to 15, continuous and intermittent data is collected. The data sources include dietary and activity logging, continuous heart rate tracking, continuous glucose monitoring, intermittent standardized meals, and intermittent low-intensity exercise. Such comprehensive and multi-dimensional data acquisition facilitates accurate modeling and prediction of glucose response, thereby enabling generation of personalized health recommendations.

[0050] A. Collection of data: A sample of adults (also referred to as patients) (for e.g. 1000 adults) with T2DM and suboptimal disease control are selected from various locations (for e.g. 14 locations) in geographically distinct regions across India, with certain inclusion and exclusion criteria, provided hereinbelow:

[0051] Table 1: Patient inclusion and exclusion criteria:

[0052] Activities performed by the patients during the period of collection of data:

[0053] Patients are fitted with a continuous glucose monitors (CGM) sensor on their upper, non-dominant arm and are provided with heart rate monitor and a glucometer with testing supplies. Patients also checked their capillary glucose on certain days before breakfast and dinner. Patients are required to consume protocol-specified meals and to perform light activity, as described in Table 2. Where applicable, patients are given several options as to which of their usual foods are acceptable for each protocol-specified food modification.

[0054] Patients are instructed to:

[0055] (a) fast for a minimum of 8 hours prior to and 3 hours after consuming the protocol-specified breakfast meal. (b) during these fasting periods, limit exercise and drink only still (not sparkling) water, tea, or coffee in moderation.

[0056] (c) eat the meal, in its entirety, within 20 minutes.

[0057] After completing the post-meal fasting period on standardized test meal days, patients can consume other foods as they normally would unless there were other meal modifications specified by the protocol later the same day.

[0058] On other days, patients are asked to consume normal foods with protocol- specified constraints. For example, on different days, patients varied the types of mixed protein (e.g., different types of lentils, with or without added protein), decided the sequence in which the foods were to be consumed, consumed of water before their meal, went for a walk after eating or eat what they perceive to be a healthy meal. If a standardized test meal is not consumed as intended, patients are provided with the option to repeat the meal.

[0059] Patients, on the smartphone, log their full dietary intake including all standardized test meals and free-living foods (including snacks), beverages (including water), medications and dietary supplements. To avoid any contingency, patients are also given a paper dietary logbook to collect the data.

[0060] Nutritional composition of standardized test meals is described in Table 3.

[0061] TABLE 3. nutritional composition of standafdized test meals

[0062] On the last day, patients were asked to remove their CGM and the endline surveys were completed to conclude the recordal of data. B. Calculation of Post Prandial Glucose Response of the patients:

[0063] Filtration of the meals data:

[0064] • Meals logged less than 30 minutes apart are merged.

[0065] • Meals logged within 90 min of other meals are removed.

[0066] • Very small (<15 g and <70 calories) meals and meals with very large (>1 kg) components are removed. • Meals with incomplete logging are removed.

[0067] • Meals consumed at the first and last 12 hour of the CGM connection are removed.

[0068] • Meals that had incomplete glucose measurements in the time window of 30 minutes before and 2 hours after the logged mealtime are removed.

[0069] The median of all glucose values from the 30-minute period prior to the meal are taken as the initial glucose level, above which the incremental area under the curve (iAUC) is calculated based on the Wolever and Jenkins method. After the consumption of meal, glucose levels typically rise as the body digests and absorbs the nutrients. The iAUC is a way to quantify the total increase in glucose levels over a period of time, usually a few hours, after consuming a meal.

[0070] The blood glucose response to each of the patients is considered for each food type. The graph below shows an example of the blood sugar response. Using the starting point as the baseline, the area under the curve is represented in gray.

[0071] Example: Food consumed and its effect on the Blood Glucose Response in a patient

[0072] C. Utilization of k-means clustering models to classify food types:

[0073] Based on the data obtained from each individual such as individual’s geographical location, age, gender, cultural background, further insights can be derived regarding behavioral patterns, preferences, health conditions, and decision-making tendencies. Such data can also help in understanding demographic diversity, identifying region-specific needs, tailoring personalized solutions, and enhancing the effectiveness of predictive models and targeted interventions. Thus, based on the analysis of responses obtained from plurality of individuals included in the predefined set of data one or more clusters are generated.

[0074] All the patients are allowed to consume protocol-specified food items. Therefore, the following nutritional information of each of the food types consumed by the patients was ascertained: protein (grams) fat (grams) carbohydrates (grams) fiber (grams) sugar (grams) total energy (kcal) time of day the food is being eaten.

[0075] Taking nutritional information of the food types consumed by the patients as an input variable, unsupervised k-means clustering models through multiple iterations, are used to classify all the foods consumed by the patients into distinct food categories.

[0076] The three distinct food categories are: High Sugar Food Normal Food High Protein Food After the classification of different food item into three distinct categories based on nutritional information, for any food item, a Gaussian mixture model uses the nutritional information of the food item to compute the overlap with the distinct food categories. For example: A food item with nutrients [Protein = 20.4g, Fat=13.4g, Carb=80.2g, Fiber=6.3g, Sugar=35.8g] would be broken down into 0.4 * High Sugar Food + 0.2 * Normal Food + 0.4 * High Protein Food.

[0077] The computed overlap makes it possible to represent the food item as a weighted linear combination of foods from the distinct food categories.

[0078] D. Classification of patients:

[0079] Each patient is classified into (a) High Class and (b) Low Class for each of the distinct food categories based on their blood glucose response to the foods in a particular distinct food category. The classes are defined using k-means clustering of the food nutrition information with blood sugar responses.

[0080] E. Training of supervised learning model to identify the Classification of patients for each of the distinct food categories:

[0081] Using the input variables from each of the individual patients, a supervised learning model is trained. Input variables from each of the patients:

[0082] 1. age (years)

[0083] 2. HbAlC (%)

[0084] 3. weight (kilos)

[0085] 4. height (inches) 5. waist (cm)

[0086] 6. hip (cm) 7. calf (cm)

[0087] 8. exercise days (days since last exercise)

[0088] 9. meals out (count per week)

[0089] 10. sitting time (min per day) 11.diabetes years (years as a diabetic)

[0090] 12. sleep duration (mean hours per night)

[0091] Training this supervised model helps to identify the high and low classifications for each of the distinct food categories for a given patient with type 2 diabetes. After the categorization of the distinct food categories and classification of patients, supervised XGBoost models (based on stochastic gradient boosting regression (such as XGBoost, version 0.6) using the XGB Regressor class, having machine learning capabilities, are trained to estimate blood sugar response for each distinct food category falling either in High Class or Low Class.

[0092] Postprandial glycemic responses are predicted as the sum of predictions from thousands of decision trees. Decision trees are inferred sequentially, with each trained on the residual of all previous decision trees and making a small contribution to the overall prediction. The features incorporated in each tree are selected by an inference procedure from a pool of features representing distinct food categories and classification of patients.

[0093] Given three distinct food categories, 6 prediction models can at least be created, each one predicting the blood sugar response (in mg / dL) for a given category of food for a given patient who either exhibits a high or low response to that category of food.

[0094] F. Final prediction of the post prandial blood glucose response: Summing up the estimated responses using the weights given to the food types, blood sugar response in any given patient is accurately predicted as a numeric output with units of milligrams per deciliter (mg / dL).

[0095] Example: Given a food with nutrition values [Protein = 20.4g, Fat=13.4g, Carb=80.2g, Fiber=6.3g, Sugar=35.8g] and a patient with T2DM having the phenotypic profile [Age=45, hbAlC=9.0, Weight=70 kg, Height=66.1 inches, Waist=88.9 cm, Hip=101.6 cm, Calf=33.0 cm, Exercise Days=2, Meals eaten out per week=3, Sitting time per day=7 hours, T2DM since=5 years, Sleep duration per day=6 hours], let us suppose we wish to estimate the blood sugar response at lunch time (Time = 1pm).

[0096] (i) the food type is broken into 0.4 * High Sugar + 0.2 * Normal + 0.4 * High Protein,

[0097] (ii) using the supervised learning model, the patient is identified to be put into either High Class or Low Class (based on his / her phenotypic profile). (iii) XGBoost model is used to estimate the blood glucose responses as (73.4, 32.5, 21.1) mg / dL for each distinct food.

[0098] (iv) The responses are summed up with appropriate weights of the food types making it possible to accurately predict the post-prandial glucose response i.e. 0.4 * 73.4 + 0.2 * 32.5 + 0.4 * 21.1 = 44.3 mg / dL.

[0099] G. Personalized recommendations to reduce blood sugar based on the predicted blood glucose response:

[0100] The present disclosure further relates to a method to provide personalized recommendations to reduce blood sugar in optimal ways by searching through all combinations of actions to rank blood sugar reducing actions for a given food and individual considering the factors including but not limited to meal type (breakfast, lunch, snack or dinner), individual meal preferences (veg, non-veg), and regional meal preferences (south Indian, north Indian) to select the best actions to reduce the blood sugar impact. The output can be a list of suggested actions that may be taken by the patient to reduce the blood sugar impact of the food being consumed, along with the estimated reduction in blood sugar impact expected by each action.

[0101] In order to ensure practicality of the method, the optimization process considers the difficulty of the actions and the effect on the taste or other characteristics of the food. This has been achieved by creating a set of complex rules within this method such as selecting different types of actions (for e.g., not all actions can be food substitution), appropriate substitutions (for e.g., swap masala dosa for plain dosa but not pav bhaji), and size of impact (for e.g., if taking a walk after having dinner is more beneficial, it should be preferred instead of taking a walk before having dinner).

[0102] For example: As part of the clinical study protocol, patients were asked to follow specified variations including food substitution (for example, swapping white rice for brown rice), food ordering (for example, eat protein before eating carbohydrates), food timing (for example, eat cooled white rice, not hot), exercise intervention (for example, walk 15 min before or after food), and food modifications (for example, make pau bhaji without potatoes). Data from these specific experiments showed statistically significant differences in blood sugar responses based on the type of action taken.

[0103] The present invention enables individuals with T2DM to understand and comprehend how their dietary choices affect their condition in real time. They can then select from a range of personalized recommendations to lessen the impact of the blood glucose on their health. This personalized approach therefore enhances the practicality of diabetes management. Benefit of the present invention lies in providing a comprehensive and personalized system for predicting post-prandial glucose response (PPGR) and delivering actionable recommendations to reduce blood sugar in individuals, particularly those with type 2 diabetes. By integrating clinical information, nutritional information, biometric measurements, laboratory data, and lifestyle parameters with trained artificial intelligence models, the invention enables highly accurate, individualized predictions of glucose response to various food items and activities. This holistic approach not only improves the precision of dietary and lifestyle recommendations but also reduces reliance on generalized treatment protocols, thereby enhancing patient compliance, minimizing risks of hyperglycemia, and improving overall metabolic health outcomes.

[0104] The present invention can be implemented through software, firmware, hardware, or any suitable combination thereof. The implementation may be organized into one or more functional components or modules, including routines, objects, programs, and data structures, each designed to perform specific tasks or abstract data types. The organization and arrangement of these modules is not limited to any particular structure illustrated in the figures or described herein; rather, the invention covers any arrangement of modules and components with more, fewer, or alternative functionalities, provided they perform the intended functions. Execution of such modules transforms a general-purpose computer into a special-purpose computing device uniquely configured to carry out the disclosed functions. The sequence of operations disclosed herein is not limiting unless explicitly specified, and operations may be performed before, concurrently with, or after one another. In certain cases, additional operations may be included, or some may be omitted, while remaining within the scope of the invention. The term “processor,” as employed herein, broadly refers to any computational unit, including single-core processors, single processors with multithread execution, multi-core processors, multi-core processors with hardware and / or software-based multithreading, parallel architectures, and distributed parallel platforms. It also includes integrated circuits, applicationspecific integrated circuits (ASICs), digital signal processors (DSPs), field- programmable gate arrays (FPGAs), programmable logic controllers (PLCs), complex programmable logic devices (CPLDs), discrete gate or transistor logic, discrete hardware components, or combinations thereof. Processors may also employ nanoscale architectures such as molecular or quantum-dot based transistors, switches, or gates to enhance efficiency, performance, or compactness. A processor may further be realized as a heterogeneous combination of multiple computing units configured to execute specialized operations.

[0105] Memory components or computer-readable storage media of the invention may comprise volatile or nonvolatile storage, or a combination thereof. Nonvolatile memory includes, but is not limited to, read-only memory (ROM), programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), and flash memory. Volatile memory includes random access memory (RAM), such as synchronous RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), and direct Rambus DRAM (DRDRAM). The invention is not limited to these specific forms but extends to any suitable memory that provides the intended functions.

[0106] When introducing elements of the invention, the articles “a,” “an,” “the,” and “said” are intended to signify one or more of the referenced elements. The terms “comprising,” “including,” and “having” are to be interpreted in an inclusive manner, permitting the presence of additional elements not expressly recited. The term “exemplary” refers to an example or instance and is not intended to imply a preferred embodiment. Similarly, the phrase “one or more of the following: A, B, and C” should be understood to include at least one of A, and / or at least one of B, and / or at least one of C. The invention as described herein is to be interpreted as illustrative and not limiting, such that variations, substitutions, and modifications may be made without departing from its broader scope. The specific features and operations disclosed are examples of implementing the claimed subject matter, and equivalent alternatives are contemplated within the scope of the appended claims.

Claims

AMENDED CLAIMS received by the International Bureau on 29 January 2026 (29.01.2026)I / We Claim:

1. A system (200) for predicting post-prandial glucose response (PPGR) in an individual with type 2 diabetes, comprising: a processing unit (204) coupled with a memory unit (202) configured to store a predefined set of data, wherein the processor is configured to perform the following steps: retrieving the predefined set of data from the memory unit (202), nutrition information of at least a food item, and phenotype information associated with at least an individual, said phenotype information associated with an individual comprises anthropometric measurements, activity level, and disease duration such as age years as a diabetic, weight, waist size, hip size, calf size, exercise days since last exercise, meal count per week, sitting time, years as a diabetic and sleep duration; configuring the phenotype information associated with said individual, and the nutrition information with separate Artificial Intelligence (Al) models, wherein each model configured to analyze said phenotype and nutritional information; classifying said phenotype information associated with an individual into one of plurality of clusters using an Al Grouping Model (206); classifying said nutritional information into distinct clusters using a Nutrition Al Model (208); said phenotype information in combination with nutritional information is configured to predict post-prandial glucose response(PPGR) with a defined granularity using a supervised Al model (214), wherein the granularity is represented on a scale of values; and said predicted postprandial glucose response (PPGR) output is further configured to generate personalized recommendations (218) to reduce blood sugar impact of food item on an individual, said personalized recommendations to reduce blood sugar impact of food item on an individual comprises:- identifying a plurality of actions associated with reducing the predicted blood glucose response;- ranking the plurality of actions based on factors including at least one of meal type, individual meal preferences, and regional meal preferences;- generating, based on said ranking, a list of recommended actions with an estimated reduction in blood sugar impact for each action.

2. The system (200) as claimed in claim 1, wherein said Nutrition Al Model (208) for nutritional information is an unsupervised machine learning model, including but not limited to clustering models such as K -Means.

3. The system (200) as claimed in claim 1 , wherein said unsupervised model classifies nutritional information into distinct categories including High Sugar, Normal, and High Protein.

4. The system (200) as claimed in claim 1 , wherein Al Grouping Model (206) for classifying clinical information comprises a supervised machine learning model.

5. The system (200) as claimed in claim 4, wherein said Al Grouping Model (206) is configured to classify an individual into most relevant phenotype cluster among a plurality of distinct phenotype clusters.

6. The system (200) as claimed in claim 4, wherein said Al Grouping Model (206) is configured to perform clustering based at least on historical blood sugar response data.

7. The system (200) as claimed in claim 1, wherein said nutritional information of an individual comprises at least one of protein, fat, carbohydrates, fiber, sugar and total calories.

8. The system (200) as claimed in claim 1, wherein said supervised Al model (216) for predicting postprandial glucose response comprises a gradient boosting model, including but not limited to an XGBoost model.

9. The system (200) as claimed in claim 1, wherein said distinct clusters of nutritional information stored in memory are derived from a clinical study quantifying postprandial glucose responses (PPGR) for Indian-specific meals.

10. The system (200) as claimed in claim 1, wherein an output module (216) is configured to present the predicted PPGR to the individual on a graphical user interface.11.The system (200) as claimed in claim 1, wherein said supervised Al model (214) is configured to predict the post-prandial glucose response (PPGR) on a graded scale of 1 to 5.

12. The system (200) as claimed in claim 1, wherein action comprises substituting one food item with another or performing a physical activity after consumption, and wherein the system applies an optimization rule to prioritize actions with higher reduction impact.

13. The system (200) as claimed in claim 1, wherein the ranked actions include substituting carbohydrate modification (brown rice for white rice), meal sequencing (delaying carbohydrate consumption until after protein intake), or exercise timing (performing post-meal walking).

14. A method (100) for predicting post-prandial glucose response (PPGR) in an individual with type 2 diabetes, the method comprising: a. retrieving (102), by a processing unit from a memory, a predefined set of data comprising phenotype information of at least one individual and nutrition information of food item, said phenotype information associated with an individual comprises anthropometric measurements, activity level, and disease duration such as age years as a diabetic, weight, waist size, hip size, calf size, exercise days since last exercise, meal count per week, sitting time, years as a diabetic and sleep duration; b. processing (104) the phenotype information with an Al Grouping Model to analyze said phenotype information; c. processing (106) the nutrition information with a Nutrition Al Model to analyze said nutrition information;d. classifying (108), by said Al Grouping Model, the individual into one of a plurality of phenotype clusters; e. classifying (110), by said Nutrition Al Model, the nutrition information into distinct nutritional clusters; f. predicting (112), by using a supervised Al model, a post-prandial glucose response (PPGR) of the individual with a defined granularity on a graded scale of 1 to 5, based on said phenotype cluster and said nutritional cluster; and g- generating (114) personalized recommendations to reduce blood sugar impact of the food item for the individual, based on the predicted PPGR, said personalized recommendations to reduce blood sugar impact of food item on an individual comprises:- identifying a plurality of actions associated with reducing the predicted blood glucose response;- ranking the plurality of actions based on factors including at least one of meal type, individual meal preferences, and regional meal preferences;- generating, based on said ranking, a list of recommended actions with an estimated reduction in blood sugar impact for each action.

Images