An artificial intelligence (AI)-based empirical antibiotic recommendation system (EARS) and method of operating the same

Abstract

The present disclosure introduces an AI-based empirical antibiotic recommendation system (100) designed to analyse patient data and recommend antibiotic treatments. The system (100) integrates with user devices via REST APIs and ensures secure operation through HTTPS. Client modules collect and pre-process raw data, ensuring integrity and compatibility. The pre-processed data is sent to the server (104) and forwarded to an input module (112) for authentication and validation. A processing module (114) standardizes the data, while a prediction module (116) identifies probable organisms and their antibiotic sensitivities. A clinical pathway module (118) generates dosing guidelines, and a prediction response module (120) displays the top three organisms with associated probabilities and recommended antibiotics. The antibiotics are color-coded for safety, and a feedback mechanism allows continuous system refinement.

Description

4. DESCRIPTION:Field of the invention:

[0001] The present disclosure generally relates to the technical field of artificial intelligence (Al) and machine learning (ML), and in specific relates to, an Al-based Empirical antibiotic recommendation system that analyses patient's data to identify one or more probable organisms and their corresponding antibiotic sensitivities for providing recommendations for Antibiotic Recommendation.Background of the invention:

[0002] Global human antibiotic consumption has skyrocketed over the past decade, particularly in middle-income countries. The use of antibiotics leads to antibiotic resistance (ABR). In specific, antibiotic resistance (ABR) is defined as the ability of bacteria to grow and adapt in the presence of antibiotics. There is a direct association between antibiotic consumption and the emergence of antibiotic resistance (ABR). Inappropriate and excessive prescribing of antibiotics contributes to the spread of ABR.

[0003] ABR has earned its place as a top ten global threat, linked to millions of deaths annually and projected to worsen in the coming decades without drastic intervention. Current interventions, like antibiotic stewardship programs and clinical decision support systems, primarily target healthcare practitioners, aiming to shift prescribing practices. While beneficial, these methods often falter, either requiring sustained effort for minimal improvement or proving difficult to integrate into workflows.

[0004] An artificial intelligence (Al) and its subfield, machine learning (ML). This technology offers a ground-breaking approach to tackling ABR. By analyzing vast datasets, machine learning (ML) algorithms can predict the best antibiotic choices, explore effective combinations, and discover new antimicrobial peptides. Supervised learning, a specific ML technique, uses past examples to make accurate predictions, paving the way for smarter, more targeted antibiotic use.2

[0005] The battle against ABR requires a multi-pronged approach. While existing methods remain vital, the potential of artificial intelligence (Al) or machine learning (ML) holds immense promise. Harnessing the power of computational analysis to guide antibiotic treatment could become a game-changer in this critical fight, preserving the life-saving power of these essential medications for generations to come.

[0006] Existing methods to determine antibiotic sensitivity, such as Antibiotic resistance testing, which involves culturing bacterial Isolates and assessing their response to various antibiotics, and demographic data are conventional methods used to predict resistance. However, these methods are time-consuming and lack precision. Machine learning algorithms like logistic regression, random forests, and neural networks have been used to predict antibiotic susceptibility using factors like patient demographics and culture data, but often fall short in personalizing recommendations.

[0007] Challenges in this field include obtaining high-quality diverse data, generalizing models across different patient populations and pathogens, and incorporating clinical context. Therefore, there is a need for more sophisticated, patient-specific approaches like Apollo's Empirical Antibiotic Recommendation System (EARS) to guide antibiotic selection. EARS employs a comprehensive data set gathered from ethically sourced data over 260,000 isolates from more than 20 locations of Apollo Hospitals across India from 2018 to 2023. The data spans all genders and age groups from 0 to 90+ years, and includes over 260,000 isolates, 57 specimen types, 181 organisms, and 152 antibiotics.

[0008] The system uses a three-layer model to predict the most likely organisms and suitable antibiotics for each case. It first recommends the top three probable organisms and their corresponding antibiotic sensitivity. It then calculates the probabilities of these organisms using a Bayesian Framework and generates a secondary output. Finally, it applies a Gradient Boosting Algorithm on the secondary output data to estimate the model's accuracy. The system has shown 92% accuracy in recommending the top three organisms and antibiotics.

[0009] While the EARS offers a sophisticated approach to antibiotic selection, it does have certain limitations, and currently does not provide recommendations on Antiviral Therapy including HAART, Anti-Malarial or Parasitic Infections Therapy, Anti Mycotic or Anti-Fungal Therapy, Anti TB Medications. Moreover, specialty-specific recommendations are in the development phase and are not yet integrated into the model. For specific syndromic approach, it is advised to refer to ICMR Guidelines as the EARS model does not recommend Empirical Antibiotics in this category yet. Nevertheless, the EARS serves as a valuable guide for clinicians, aiding them in selecting the most effective empirical antibiotic therapy. This not only enhances patient outcomes but also contributes to a reduction in the misuse of antibiotics.Objectives of the invention:

[0010] The primary objective of the invention is to provide an Al-based system capable of analysing patient data to ascertain probable organisms and corresponding antibiotic sensitivities, thereby offering Antibiotic treatment recommendations.

[0011] Another objective of the invention is to provide an Al-based system to limit the use of broad-spectrum antibiotics by providing personalized recommendations, hence promoting more targeted and efficient treatment.

[0012] The other objective of the invention is to provide an Al-based system that improves the accuracy of antibiotic selection by considering the patient's data.

[0013] The other objective of the invention is to enhance the precision of antibiotic selection by taking into account comprehensive patient data.

[0014] The other objective of the invention is to optimize antibiotic usage based on specific infections and patient profiles, thus personalizing treatment plans.

[0015] Yet another objective of the invention is to identify global risks for antibiotic resistance by analysing patient data, contributing to a more informed approach to antibiotic use.

[0016] Further objective of the invention is to provide an Al-based system that promotes collaboration between artificial intelligence (Al) and clinicians to assist in decision-making and improving patient outcomes.Summary of the invention:

[0017] The present disclosure proposes an artificial intelligence (Al)-based empirical antibiotic recommendation system (EARS) and method thereof. The following presents a simplified summary in order to provide a basic understanding of some aspects of the claimed subject matter. This summary is not an extensive overview. It is not intended to identify key / critical elements or to delineate the scope of the claimed subject matter. Its sole purpose is to present some concepts in a simplified form as a prelude to the more detailed description that is presented later.

[0018] In order to overcome the above deficiencies of the prior art, the present disclosure is to solve the technical problem to provide an Al-based system that provides guidance on empirical antibiotic therapy selection, considering patient characteristics, infection site, and local resistance patterns.

[0019] According to an aspect, the invention provides an artificial intelligence (Al)-based empirical antibiotic recommendation system (EARS). In one embodiment herein, the server having a processor and a memory for storing one or more instructions executable by the processor. Additionally, the processor is configured to execute plurality of modules for identifying and recommending probable organisms and their sensitive antibiotics at the point of care. The server is in communication with a user device through a network. The server is in communication with a database for storing and retrieving the patient's data.

[0020] In one embodiment herein, the user device includes at least one of a computer, smartphone, and laptop. The system supports integration with a user device through various approaches, including the use of REST APIs. This integration method ensures seamless operation and generally employs HTTPS for secure communication.

[0021] The system features a server connected to a network and communicates with a database for managing data efficiently. The server is equipped with a processor and memory, with the processor running various modules to identify and recommend probable organisms and their sensitive antibiotics at the point of care. The processor is configured to execute plurality of modules for identifying and recommending probable organisms and their sensitive antibiotics at the point of care. The plurality of modules comprises client modules, an input module, a processing module, a prediction module, a clinical pathway module and a prediction response module.

[0022] In one embodiment herein, the client modules is configured to perform multiple functions identifying and recommending probable organisms and their sensitive antibiotics at the point of care. The client modules comprise a data collection module, a standardization and pre-processing module and an API module. In one embodiment herein, the data collection module is configured to collect input data from various sources such as electronic medical records (EMRs), Data repositories and Databases.

[0023] In one embodiment herein, the standardization and pre-processing module is configured to standardize and pre-process the collected data to ensure its integrity and compatibility with the system's processing requirements. In one embodiment herein, the API module is configured to transmit the pre-processed data from the data collection module and the standardization and pre-processing module to the server via a network.

[0024] In one embodiment herein, the input module is responsible for authenticating and validating data inputs, thereby ensuring they meet the required formats. It collects comprehensive data, including personal and demographic details, clinical conditions, service type, specimen type, liver and kidney function, hypersensitivity to antibiotics, and antibiotic use history, and then sends this data to the processing module for standardization.

[0025] In one embodiment herein, the processing module transforms and standardizes data for subsequent analysis. In one embodiment herein, the prediction module analyzes the individual's profile against a comprehensive dataset to predict the top three organisms andtheir sensitivity to antibiotics. This module uses multiple parameters to ensure maximum efficiency in its predictions.

[0026] In one embodiment herein, the clinical pathway module generates a recommended protocol for subsequent actions, including standard dosing guidelines for both adults and pediatric populations. The prediction response module then presents these predictions, color-coded for clarity, to assist clinicians in making informed decisions. In one embodiment herein, the system processes patient data, which includes personal details such as name, age, gender, and pregnancy status, as well as demographic details, clinical condition, service type, specimen type, liver and kidney function, hypersensitivity to antibiotics, and antibiotic use history.

[0027] The predictive algorithms in the system are organized into three layers such a first layer, a second layer and a third layer. The first layer uses content-based filtering to identify likely organisms and their sensitivities. The second layer applies a Bayesian framework to calculate probabilities, creating secondary output data. The third layer utilizes a gradient boosting algorithm to assess model accuracy.

[0028] In one embodiment herein, the system utilizes a comprehensive dataset consisting of over 260,000 isolates, 57 specimen types, 181 organisms, and 152 antibiotics collected from more than 20 locations across India. This extensive dataset supports the accuracy of predictions. Additionally, the prediction response module incorporates a feedback mechanism that allows users to provide input on the recommended antibiotics. This feedback is used to continuously improve the system's accuracy and performance.

[0029] According to another aspect, the invention provides a method for operating the artificial intelligence (Al)-based empirical antibiotic recommendation system (EARS). At one step, the data collection module collects the input data from various sources such as electronic medical records (EMRs), Data repositories and Databases. At one step, the standardization and pre-processing module standardizes and pre-processes the collected data to ensure its integrity and compatibility with the system's processing requirements.

[0030] At one step, the API module transmit the pre-processed data from the data collection module and the standardization and pre-processing module to the server via the network. At one step, the input module receives the pre-processed data from the server to authenticate and validate, and collects patient-related parameters and medical history, thereby transmitting the input data to a processing module for further processing. At one step, the processing module standardizes the received data to fit the requirements for further analysis. At one step, the prediction module receives the processed data from the processing module and provides an output.

[0031] At one step, the clinical pathway module generates the recommended protocol of subsequent actions, such as standard dosing guidelines, based on the proposed antibiotics. At one step, the prediction response module presents the top three most likely organisms and their sensitivity patterns, along with location-based results that are color-coded for clarity, and include a feedback mechanism for continuous system improvement.

[0032] Further, objects and advantages of the present invention will be apparent from a study of the following portion of the specification, the claims, and the attached drawings.Detailed description of drawings:

[0033] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate an embodiment of the invention, and, together with the description, explain the principles of the invention.

[0034] FIG. 1 illustrates a block diagram of an artificial intelligence (Al)-based empirical antibiotic recommendation system (EARS), in accordance to an exemplary embodiment of the invention.

[0035] FIG. 2 illustrates a process of an artificial intelligence (Al)-based empirical antibiotic recommendation system (EARS), in accordance to an exemplary embodiment of the invention.

[0036] FIGs. 3A-3B illustrate graphical representations of AUC ROC and AUC PRC Curves of all specimen obtained from an artificial intelligence (Al)-based empirical antibiotic recommendation system (EARS), in accordance to an exemplary embodiment of the invention.

[0037] FIGs. 4A-4B illustrate graphical representations of AUC ROC and AUC PRC Curves of urine culture obtained from an artificial intelligence (Al)-based empiric antibiotic recommendation system (EARS), in accordance to an exemplary embodiment of the invention.

[0038] FIGs. 5A-5B illustrate graphical representations of AUC ROC and AUC PRC Curves of pus and wound swab obtained from an artificial intelligence (Al)-based empirical antibiotic recommendation system (EARS), in accordance to an exemplary embodiment of the invention.

[0039] FIGs. 6A-6B illustrate graphical representations of AUC ROC and AUC PRC Curves of blood culture obtained from an artificial intelligence (Al)-based empirical antibiotic recommendation system (EARS), in accordance to an exemplary embodiment of the invention.

[0040] FIGs. 7A-7B illustrate graphical representations of AUC ROC and AUC PRC Curves of respiratory specimen culture obtained from an artificial intelligence (Al)-based empirical antibiotic recommendation system (EARS), in accordance to an exemplary embodiment of the invention.

[0041] FIGs. 8A-8B illustrate graphical representations of AUC ROC and AUC PRC Curves of stool culture obtained from an artificial intelligence (Al)-based empirical antibiotic recommendation system (EARS), in accordance to an exemplary embodiment of the invention.9

[0042] FIG. 9 illustrates a flowchart of a method for operating an artificial intelligence (Al)- based empirical antibiotic recommendation system (EARS), in accordance to an exemplary embodiment of the invention.

[0043] FIG. 10 illustrates a system architecture (1000) for developing and deploying an artificial intelligence (Al)-based empirical antibiotic recommendation system (EARS), in accordance to an exemplary embodiment of the invention.Detailed invention disclosure:

[0044] Various embodiments of the present invention will be described in reference to the accompanying drawings. Wherever possible, same or similar reference numerals are used in the drawings and the description to refer to the same or like parts or steps.

[0045] The present disclosure has been made with a view towards solving the problem with the prior art described above, and it is an object of the present invention to provide an Al- based system 100 that analyses patient's data to identify one or more probable organisms and their corresponding antibiotic sensitivities for providing recommendations for antibiotic prescription.

[0046] The empirical antibiotic recommendation system is an advanced Al tool that provides trends in antibiotic use based on age, gender, patient status (inpatient or outpatient), infection site, and days since admission. Utilizing data from 2018 to 2023, it employs a content-based machine learning recommendation system, analysing 33% outpatient and 66% inpatient isolates. The model accurately predicts the top three probable organisms and their sensitivity patterns with over 92% accuracy. Updated quarterly, it incorporates data from over 20 Apollo Hospitals, following WHONET and AWARE classifications from the WHO.

[0047] According to an exemplary embodiment of the invention, FIG. 1 refers to a block diagram of an artificial intelligence (Al)-based empirical antibiotic recommendation system (EARS) 100. In one embodiment, the system 100 supports integration with a user device through various approaches, one of the methods is use of REST APIs. This integrationmethod ensures seamless operation and generally employs HTTPS for secure communication.

[0048] In one embodiment herein, the server 104 having a processor 106 and a memory 108 for storing one or more instructions executable by the processor 106. In particular, the processor 106 is configured to execute plurality of modules for identifying and recommending probable organisms and their sensitive antibiotics at the point of care. The server 104 is in communication with the user device through a network 102. The server 104 is in communication with a database 110 for storing and retrieving the patient's data. The plurality of modules comprises client modules, an input module 112, a processing module 114, a prediction module 116, a clinical pathway module 118, and a prediction response module 120.

[0049] In one embodiment herein, the client modules are configured to perform multiple functions for identifying and recommending probable organisms and their sensitive antibiotics at the point of care. The client modules comprises a data collection module 122, a standardization and pre-processing module 124 and an API module 126. In one embodiment herein, the data collection module 122 is configured collect the input data from various sources such as electronic medical records (EMRs), Data repositories and Databases.

[0050] In one embodiment herein, the standardization and pre-processing module 124 is configured to standardize and pre-processes the collected data to ensure its integrity and compatibility with the system's processing requirements. In one embodiment herein, the API module is configured to transmit the pre-processed data from the data collection module 122 and the standardization and pre-processing module 124 to the server 104 via the network.

[0051] In one embodiment, the input module 112 receives the pre-processed data form the server 104 to authenticate and validate, ensuring compliance with required formats. This module enables users to input patient's data, including but not limited to personal details such as name, age, gender, and pregnancy status if applicable. It incorporates informationabout the patient's geographical location in India, which can range from cities like Bangalore, Bhubaneswar, Chennai, Delhi, Hyderabad, Kolkata, Mumbai, Nashik, and others.

[0052] The input module 112 also inputs the current clinical condition of the patient and the service type, whether it's Outpatient (OP), Emergency Room (ER), and Inpatient (IP) and Number of days since admission if applicable. It collects data about the type of specimen provided for testing, such as BAL, Blood, Cerebrospinal Fluid, Nasal Swab, Pleural Fluid, Pus, Rectal Swab, Sputum, Stool, Throat Swab, Tracheal Secretion, Urine, or a Wound Swab. Additionally, it gathers information about the patient's liver and kidney function, requiring further details if there's any impairment, SGPT, SGOT, Bilirubin; Creatinine, Urea, Albuminuria among others values need to be provided. The input module 112 also captures data about any hypersensitivity the patient may have towards specific antibiotics and their current antibiotic use history. The collected data is transmitted to processing module for further processing.

[0053] In one embodiment, the processing module 114, transforms and standardizes data from the input module to meet the necessary input criteria for the subsequent analysis. In one embodiment, the prediction module 116, designed to predict the top three organisms and the top three antibiotics for each organism along with location-based results, analyses the individual's profile— factors like age, gender, inpatient or outpatient status, site of infection, and days since admission. It leverages a comprehensive dataset, comprising over 260K isolates, 57 specimen types, 181 organisms, and 152 antibiotics, gathered from more than 20 locations across India.

[0054] The system 100 is further optimized with location-specific prescriptions of organisms and corresponding antibiotics. The prediction module 116 is structured in three layers for maximum efficiency, the first layer employs content-based filtering to yield the top three probable organisms along with their corresponding antibiotic sensitivity. The second layer uses a Bayesian framework to calculate each of these three organisms' probabilities, subsequently creating a secondary output data through the summation of the top three recommended organisms' probabilities.

[0055] The metrics results, including cosine similarity between the original organism and the top three predicted organisms, are then computed. The third layer applies a gradient boosting algorithm on the secondary output data to estimate the model's accuracy, including multiple sub-models. In one embodiment, the clinical pathway 118 module is designed to provide a recommended protocol of subsequent actions for at least one Individual, based on the recommended Antibiotics.

[0056] In one embodiment, the prediction response module 120 unveils the three most likely organisms along with their respective top three antibiotics, determined by their sensitivity patterns. Additionally, it also highlights the top two organisms and their sensitivity patterns in relation to the individual's location. To enhance the clarity of information, antibiotics are presented with specific colour codes. Grey signifies the ongoing administration of the antibiotic to the patient; yellow warns of potential complications with liver diseases or impaired liver enzymes.

[0057] Pink denotes caution due to kidney diseases and increased creatinine and urea levels; red signifies known hypersensitivity to the specific antibiotic or those within the same group; blue suggests potential ineffectiveness on certain organisms due to inherited resistance or the mechanism of action; orange denotes antibiotics that may not target certain organ systems, recommending that the specific antibiotic-organism combination be appropriately verified. In the case of antibiotics coded in Teal or sky blue, it's suggested that their usage be cautiously considered during pregnancy or breastfeeding.

[0058] The prediction response module 120 also presents feedback mechanism that enables the at least one user to provide feedback on the recommended antibiotic, which is used to continuously improve the system's accuracy and performance. In one embodiment herein, the user device includes a computer, a smartphone, and a laptop.

[0059] According to another embodiment of the invention, FIG. 2 refers to a process of an artificial intelligence (Al)-based empiric antibiotic recommendation system (EARS) for antibiotic treatment. In one embodiment herein, the artificial intelligence (Al)-based empiric antibiotic recommendation system (EARS) 100 considers multiple factors, which includepatient demographics, clinical information, and laboratory results. The multiple factors are used for recommending appropriate antibiotic treatment options.

[0060] In one embodiment herein, the process of the artificial intelligence (Al)-based empiric antibiotic recommendation system (EARS) comprises a primary phase, a secondary phase and a tertiary phase. In one embodiment herein, the primary phase involves in collecting patient's data from multiple sources such as Electronic Medical records and laboratory databases. In specific, the patient's data includes such as name, age, gender, and pregnancy status and demographic Details, Current clinical condition, service type (such as outpatient, emergency room, or inpatient, including the number of days since admission if inpatient), specimen type, liver and kidney function, hypersensitivity to antibiotics, and antibiotic use.

[0061] The machine learning techniques currently used in three layers include contentbased filtering, Bayesian Framework, and extreme Gradient Boost. The first layer employs content-based filtering to yield the top three probable organisms along with their corresponding antibiotic sensitivity. The second layer uses a Bayesian Framework to calculate each of these three organisms' probabilities, subsequently creating a secondary output data through the summation of the top three recommended organisms' probabilities.

[0062] The metrics results, including Cosine Similarity between the original organism and the top three predicted organisms, are then computed. The third layer applies a Gradient Boosting Algorithm on the secondary output data to estimate the model's accuracy, including multiple sub-models, using a 70-30 train-test data divide.

[0063] In one embodiment, the secondary phase enhances the system by developing a machine learning model that detects data patterns and predicts antibiotic susceptibility, utilizing lab results, imaging, and discharge summaries. In another embodiment, the tertiary phase aims to integrate this model, enriched with Bayesian structure learning, into a clinical decision support system to assist healthcare providers in making informed treatment decisions. In the future, we plan to further develop and refine this system to enhance itspredictive capabilities and expand its integration with diverse clinical data sources and decision-making frameworks.

[0064] According to another embodiment of the invention, FIGs.3A-3B refer to graphical representations (300,302) of Area Under the Receiver Operating Characteristic curve (AUC- ROC), Precision Recall (PR) Curve (300,302) of all specimen obtained from an artificial intelligence (Al)-based empirical antibiotic recommendation system (EARS) 100. In one embodiment herein, ROC curve is a graphical representation of the performance of a binary classification model. In specific, true positive rate (TPR) is represented on Y-axis and false positive rate (FPR) is represented on the X-axis.

[0065] The TPR is the proportion of positive cases that are correctly identified as positive by the model. The FPR is the proportion of negative cases that are incorrectly identified as positive by the model. The ROC curve in the graph 300 depicts that the model has an AUC of 0.926. The AUC is the area under the ROC curve, and it represents the probability that the model will rank a randomly chosen positive instance higher than a randomly chosen negative instance. An AUC of 1 means that the model will always rank positive instances higher than negative instances, and an AUC of 0.5 means that the model is no better than random guessing.

[0066] The ROC curve in the graph 302 also depicts that the model has an AP of 0.814. The AP is the average precision, which is a measure of how precise the model's predictions are. An AP of 1 means that all of the model's predictions are correct, and an AP of 0 means that none of the model's predictions are correct. However, the ROC curve in the graphical representations (300,302) depicts that the model is performing well. In specific, the AUC is high, and the AP is also high. This suggests that the model can correctly identify positive instances and that its predictions are precise.

[0067] According to another embodiment of the invention, FIGs. 4A-4B refer to graphical representations (400, 402) of urine culture obtained from an artificial intelligence (Al)-based empiric antibiotic recommendation system (EARS) 100. In one embodiment herein, the ROC curve is a graphical tool used to assess the performance of a binary classification model. Thegraph 400 plots the true positive rate (TPR) on the Y-axis against the false positive rate (FPR) on the X-axis.

[0068] The TPR is the proportion of positive cases that are correctly identified by the model, while the FPR is the proportion of negative cases that are incorrectly identified as positive. An ROC curve that is closer to the upper left corner of the plot indicates better performance. The ROC curve in graph 400 has an AUC of 0.92, which is considered to be a very good performance.

[0069] The X-axis of graph 402 depicts Recall and the Y-axis of graph 402 represents Precision. Recall is another measure of the performance of a binary classification model, and it is defined as the proportion of positive cases that are correctly identified by the model. Precision is a measure of the quality of the model's positive predictions, and it is defined as the proportion of predicted positive cases that are positive.

[0070] The ROC curve in graph 402 also depicts that the model has an AP of 0.876. The AP is the average precision, which is a measure of how precise the model's predictions are. An AP of 1 means that all of the model's predictions are correct, and an AP of 0 means that none of the model's predictions are correct.

[0071] According to another embodiment of the invention, FIGs. 5A-5B refer to graphical representations (500, 502) of pus and wound swab obtained from an artificial intelligence (Al)-based empiric antibiotic recommendation system (EARS). In one embodiment herein, the graph 500 depicts the Area under the Curve (AUC) of the Receiver Operating Characteristic (ROC) curve.

[0072] The ROC curve is a graphical representation of the performance of a binary classification model at all possible classification thresholds. The AUC is a measure of the overall performance of the model, and it ranges from 0 to 1. A value of 1 indicates perfect performance, while a value of 0 indicates that the model is no better than random guessing. In the image, the AUC is 0.9457, which indicates that the model has good performance.

[0073] In one embodiment herein, the X-axis of graph 502 depicts Recall and the Y-axis of graph 502 represents Precision. The Precision Recall curve in graph 502 also depicts that the model has an AP of 0.868. The AP is the average precision, which is a measure of how precise the model's predictions are. An AP of 1 means that all of the model's predictions are correct, and an AP of 0 means that none of the model's predictions are correct.

[0074] According to another embodiment of the invention, FIGs. 6A-6B illustrate graphical representations of AUC-ROC Curve, Precision Recall (PR) Curve (600, 602) of blood culture obtained from an artificial intelligence (Al)-based empiric antibiotic recommendation system (EARS). In one embodiment herein, the graph 600 depicts the Area under the Curve (AUC) of the Receiver Operating Characteristic (ROC) curve.

[0075] The ROC curve is a graphical representation of the performance of a binary classification model at all possible classification thresholds. The AUC is a measure of the overall performance of the model, and it ranges from 0 to 1. A value of 1 indicates perfect performance, while a value of 0 indicates that the model is no better than random guessing. In the image, the AUC is 0.991, which indicates that the model has good performance.

[0076] In one embodiment herein, the X-axis of graph 602 depicts Recall and the Y-axis of graph 602 represents Precision. The ROC curve in graph 602 also depicts that the model has an AP of 0.880. The AP is the average precision, which is a measure of how precise the model's predictions are. An AP of 1 means that all of the model's predictions are correct, and an AP of 0 means that none of the model's predictions are correct.

[0077] According to another embodiment of the invention, FIGs. 7A-7B refer to graphical representations of AUC-ROC Curve, Precision Recall (PR) Curve (700, 702) of respiratory specimen culture obtained from an artificial intelligence (Al)-based empiric antibiotic recommendation system (EARS). In one embodiment herein, the graph 700 depicts the Area under the Curve (AUC) of the Receiver Operating Characteristic (ROC) curve.

[0078] The ROC curve is a graphical representation of the performance of a binary classification model at all possible classification thresholds. The AUC is a measure of theoverall performance of the model, and it ranges from 0 to 1. A value of 1 indicates perfect performance, while a value of 0 indicates that the model is no better than random guessing. In the image, the AUC is 0.939, which indicates that the model has good performance.

[0079] In one embodiment herein, the X-axis of graph 702 depicts Recall and the Y-axis of graph 702 represents Precision. The Precision Recall (PR) curve in graph 702 also depicts that the model has an AP of 0.705. The AP is the average precision, which is a measure of how precise the model's predictions are. An AP of 1 means that all of the model's predictions are correct, and an AP of 0 means that none of the model's predictions are correct.

[0080] According to another embodiment of the invention, FIGs. 8A-8B refer to graphical representations of AUC-ROC Curve, Precision Recall (PR) Curve (800, 802) of stool culture obtained from an artificial intelligence (Al)-based empiric antibiotic recommendation system (EARS). In one embodiment herein, the graph 800 depicts the Area under the Curve (AUC) of the Receiver Operating Characteristic (ROC) curve.

[0081] The ROC curve is a graphical representation of the performance of a binary classification model at all possible classification thresholds. The AUC is a measure of the overall performance of the model, and it ranges from 0 to 1. A value of 1 indicates perfect performance, while a value of 0 indicates that the model is no better than random guessing. In the image, the AUC is 0.967, which indicates that the model has good performance.

[0082] In one embodiment herein, the X-axis of graph 802 depicts Recall and the Y-axis of graph 802 represents Precision. The Precision Recall (PR) curve in graph 802 also depicts that the model has an AP of 0.961. The AP is the average precision, which is a measure of how precise the model's predictions are. An AP of 1 means that all of the model's predictions are correct, and an AP of 0 means that none of the model's predictions are correct.

[0083] According to another embodiment of the invention, FIG. 9 refers to a flowchart 900 of a method for operating an artificial intelligence (Al)-based empiric antibioticrecommendation system (EARS) 100. At step 902, the data collection module 122 collects the input data from various sources such as electronic medical records (EMRs), Data repositories and Databases. At step 904, the standardization and pre-processing module 124 standardizes and pre-processes the collected data to ensure its integrity and compatibility with the system's processing requirements.

[0084] At step 906, the API module 126 transmits the pre-processed data from the data collection module 122 and the standardization and pre-processing module 124 to the server 104 via the network 102. The input module 112 receives the pre-processed data from the server 104 at step 908 to authenticate and validate. The pre-processed data includes but not limited to personal details such as name, age, gender, and pregnancy status and demographic details, current clinical condition, service type (such as outpatient, emergency room, or inpatient, including the number of days since admission if inpatient), specimen type, liver and kidney function, hypersensitivity to antibiotics, and antibiotic use history.

[0085] This data is then transmitted to processing module for further processing. At step 910, the processing module 114 standardizes and transforms the input data to meet the required criteria for subsequent analysis.

[0086] At step 912, the prediction module 116 operates in three layers to optimize efficiency. The first layer utilizes content-based filtering to identify the top three probable organisms and their corresponding antibiotic sensitivities. The second layer uses a Bayesian framework to calculate the probabilities for each of these three organisms, generating secondary output data by summing the probabilities of the top three recommended organisms. Metrics, including the cosine similarity between the original organism and the top three predicted organisms, are computed.

[0087] The third layer applies a Gradient Boosting Algorithm to this secondary output data, estimating the model’s accuracy and incorporating various sub-models. At step 914, the clinical pathway module 118 provides a recommended protocol for subsequent actions based on the suggested antibiotics for the individual. At step 916, the prediction response module 120 presents the top three likely organisms and their suitable antibiotics, taking intoaccount sensitivity patterns and the user's location. The antibiotics are color-coded to indicate their current administration status, potential complications, effectiveness, and safety during pregnancy or breastfeeding. The system also incorporates user feedback to continuously improve its accuracy and performance.

[0088] According to an exemplary embodiment of the invention, FIG. 1000 illustrates the system architecture for developing and deploying an Artificial Intelligence (Al)-based Empirical Antibiotic Recommendation System (EARS). This system architecture outlines the sequential phases integral to the system's development and deployment. In step 1002, the process begins with the data sourcing phase, where information is collected from various sources, which includes but not limited to patient demographics such as age and gender, patient status (inpatient or outpatient), infection site, and days since admission, Electronic Medical Records (EMR) 1014 and servers and clinical knowledge bases 1016.

[0089] In step 1004, we move to the data ingestion phase. Here, data is collected and put into a centralized database 1020. This data then traverses through a data pipeline 1022 and is then stored in a Data Repository 1024 in structured formats. This phase is crucial for organizing the data and preparing it for further pre-processing stages. In step 1006 includes the data pre-processing phase, which encompasses metadata management 1026, ETL processes 1028, and data transformation and harmonization 1030. These processes improve data quality and consistency, setting the stage for subsequent analysis and model training.

[0090] In step 1008, for data analysis we utilized a variety of statistical tools 1008. Propensity matching 1032 unveils hidden patterns, while Descriptive statistics 1034 provide a summarized overview of the data. Correlation coefficients 1036 establish relationships between various risk factors, and odds / hazard ratios 1038 compare the possibility of outcomes. KM plots and survival charts 1040 visualize time-to-event data, and visualization tools 1042 simplify complex data representation. These tools, in concert, contribute to feature selection and model building of the prediabetes risk assessment system.

[0091] In Step 1010, we move towards introduces a multi-layer approach to refine recommendations. The first layer uses content-based filtering to output the top threeprobable organisms along with their corresponding antibiotic sensitivities. The second layer applies a Bayesian Framework to determine and summarize the probabilities for these organisms, incorporating cosine similarity metrics for prediction accuracy. The third layer utilizes a Gradient Boosting Algorithm on the secondary output data, employing a 70-30 Train-Test data split to estimate model accuracy, with metrics such as AUROC (Area Under the Receiver Operating Characteristic Curve) and AUPRC (Area Under the Precision-Recall Curve).

[0092] In Further Step 1012 involves deployment of the model for integration and usage through REST API 1052 protocols. This architecture involves using an API Management Service 1044 and an application service resource 1050 for model inference. The model inference code is developed in Python programming language 1054. All the resources are hosted in a virtual private network 1048 securely. The API service 1046 acts as the interface for the REST API, facilitating communication between the API Management Service and the web application. The API Management Service includes storage capabilities 1058 for securely managing and storing model artifacts, data inputs, and results.

[0093] In one such application of the integration, this model inference API is integrated into a web application code 1056 to be deployed and used as a Web Application 1060. This web application serves as a user-friendly platform for users or clinicians, enabling them to input data and obtain Empiric Antibiotic Recommendations. Therefore, this phase makes the model operational and accessible, facilitating the recommendation of antibiotics empirically.

[0094] Numerous advantages of the present disclosure may be apparent from the discussion above. In accordance with the present disclosure, an artificial intelligence (Al)- based empiric antibiotic recommendation system (EARS) is disclosed. The proposed Al- based system 100 analyses patient's data and provides recommendations for antibiotic treatment. The proposed Al-based system 100 reduces the use of broad-spectrum antibiotics through personalized recommendations.

[0095] The proposed Al-based system 100 enhances antibiotic usage by optimizing antibiotic selection contributing significantly to the field of antibiotic stewardship and thefight against antimicrobial resistance. With high accuracy (92%) and recall (88%), it recommends three organisms and sensitivity pattern. It aids clinicians in choosing appropriate empirical antibiotics, reducing unnecessary broad-spectrum antibiotic use, preventing resistant infections, and improving patient outcomes.

[0096] It will readily be apparent that numerous modifications and alterations can be made to the processes described in the foregoing examples without departing from the principles underlying the invention, and all such modifications and alterations are intended to be embraced by this application.22

Claims

5. CLAIMS: l / We Claim:

1. An artificial intelligence (Al)-based empirical antibiotic recommendation system (EARS) (100), comprising: a server (104) having a processor (106) and a memory (108) for storing one or more instructions executable by the processor (106), wherein the processor (106) is configured to execute plurality of modules for identifying and recommending probable organisms and their sensitive antibiotics at the point of care, wherein the server (104) is in communication with a user device through a network (102), wherein the server (104) is in communication with a database (110) for storing and retrieving the patient's data, wherein the plurality of modules comprises: client modules configured to perform multiple functions for identifying and recommending probable organisms and their sensitive antibiotics at the point of care, wherein the client modules comprise: a data collection module (122) configured to collect input data from various sources such as electronic medical records (EMRs), Data repositories and Databases; a standardization and pre-processing module (124) configured to standardize and pre-process the collected data to ensure its integrity and compatibility with the system's processing requirements; an API module (126) configured to transmit the pre-processed data from the data collection module (122) and the standardization and pre-processing module (124) to the server (104) via a network (102); an input module (112) configured to receive the pre-processed data from the server (104) to authenticate and validate input data, ensuring compliance with required formats from various sources, and facilitating the collection of data, includes personal details, demographic details, current clinical condition, service type,23specimen type, liver and kidney function, hypersensitivity to antibiotics, and antibiotic use history, thereby transmitting the collected data to processing module; a processing module (114) configured to transform and standardize data from the input module (112) for subsequent analysis; a prediction module (116) configured to analyze an individual's profile and leverages a comprehensive dataset to predict the top three organisms and their antibiotics sensitivity pattern using multiple parameters for maximum efficiency; a clinical pathway module (118) configured to provide a recommended protocol of subsequent actions for individuals, such as standard dosing guidelines for adult and pediatric populations; and a prediction response module (120) configured to present the most likely organisms and their antibiotics with color codes for clarity, whereby the system (100) functions as a software as a medical device (SaMD), certified by ISO 13485, and integrates algorithms for recommending the top three probable organisms with their sensitivity pattern, thereby enhancing patient outcomes and supporting clinicians.

2. The Al-based EARS (100) as claimed in claim 1, wherein the patient's data includes personal details such as name, age, gender, pregnancy status and demographic details, current clinical condition, service type (outpatient, emergency room, inpatient), specimen type, liver and kidney function, hypersensitivity to antibiotics, and antibiotic use history.

3. The Al-based EARS (100) as claimed in claim 1, wherein the multiple parameters include content-based filtering, a Bayesian framework, and a gradient-boosting algorithm in three layers.

4. The Al-based EARS (100) as claimed in claim 1, wherein the prediction response module (120) includes a feedback mechanism for continuous improvement.

5. The Al-based EARS (100) as claimed in claim 1, wherein the comprehensive dataset includes over 260K isolates, 57 specimen types, 181 organisms, and 152 antibiotics gathered from more than 20 locations across India.

6. The Al-based EARS (100) as claimed in claim 1, wherein the prediction module (116) operates in three layers for optimal efficiency, wherein the three layers comprise: a first layer uses content-based filtering to identify the top three probable organisms and their antibiotic sensitivity; a second layer employs a Bayesian framework to calculate the probabilities of these organisms, creating secondary output data; and a third layer uses a gradient boosting algorithm on this data to assess the model's accuracy with multiple sub-models.

7. The Al-based EARS (100) as claimed in claim 1, wherein the Al-based EARS (100) comprises a feedback mechanism that enables the user to provide feedback on the recommended antibiotic, which is used to continuously improve the system's accuracy and performance.

8. The Al-based EARS (100) as claimed in claim 1, wherein the user device includes a computer, a smartphone, and a laptop.

9. A method for operating an artificial intelligence (Al)-based empirical antibiotic recommendation system (EARS) (100), comprising: collecting, by a data collection module (122), input data from various sources such as electronic medical records (EMRs), Data repositories and Databases; standardizing and pre-processing, by a standardization and pre-processing module (124), the collected data to ensure its integrity and compatibility with the system's processing requirements; transmitting, by an API module (126), the pre-processed data from the data collection module (122) and the standardization and pre-processing module (124) to the server (104) via a network (102);receiving, by an input module (112), the pre-processed data to authenticate and validate, and collecting patient-related parameters and medical history; standardizing, by processing module (114), the received data to fit the requirements for further analysis; receiving, by a prediction module (116), the processed data from the processing module (114) and providing an output; generating, by a clinical pathway module (118), a recommended protocol of subsequent actions, such as standard dosing guidelines, based on the proposed antibiotics; and presenting, by a prediction response module (120), the top three most likely organisms and their sensitivity patterns, along with location-based results that are color-coded for clarity, and include a feedback mechanism for continuous system improvement.

6. DATE AND SIGNATURE:Dated this 28thday of August, 2024PATENT AGENT NAME: VARALAKSHMI VANAMINPA - 32826

Images