Machine learning based system and method for identifying patients at risk of non-adherence and relevant patient intervention plans

Abstract

Disclosed herein is a healthcare management system and method utilizing machine learning based predictive analytics to improve patient adherence rates for chronic care management (e.g., diabetes). An example system may include a computing device configured to establish a customer analytical record to synthetize over a plurality of selected attributes to form a patient centric view of patient behaviors, attitude, characteristics based upon data related to chronic disease state management and social determinants of health, consumer experience to determine, using a predictive artificial intelligence model, a patient's predictive adherence for next 1-3 months, and determine tailored intervention plans to engage patients for continuing adherence.

Description

Attorney Docket No. 3726023.00009 MACHINE LEARNING BASED SYSTEM AND METHOD FOR IDENTIFYING PATIENTS AT RISK OF NON-ADHERENCE AND RELEVANT PATIENT INTERVENTION PLANSCross Reference to Related Application

[0001] The application claims priority to U.S. Provisional Patent Application No. 63 / 625,764, filed on January 26, 2024, entitled “MACHINE LEARNING BASED SYSTEM AND METHOD FOR IDENTIFYING PATIENTS AT RISK OF NON-ADHERENCE AND RELEVANT PATIENT INTERVENTION PLANS,” the content of which is incorporated by reference herein in its entirety.Field of Technology

[0002] The present disclosure generally relates to a healthcare management system and method utilizing machine learning (ML) based predictive analytics to improve patient adherence rates for chronic care management (e.g., diabetes), and more particularly relates to using advanced ML based models to understand patterns of continuous glucose monitoring / monitor (CGM) patients behavior to proactively (e.g, 3 months in advance) identify patients at risk of nonadherence in CGM therapy and to determine relevant patient intervention plans.Background

[0003] Living with any chronic conditions such as diabetes, which may lead to the potential adverse effect of hypoglycemia, is more than a full-time job. It is a 24 / 7 commitment to monitoring thousands of minute details related to diet, exercise, sleep, stress, and overall health. People living with diabetes may make nearly two hundred extra decisions every’ day to keep on top of their conditions, creating a near-constant risk that a split-second of inattention can snowball into a slide away from healthy habits and appropriate adherence to care. While diabetes adherence has historically focused on medications, modem diabetes care plan may include a number of devices such as insulin pumps and continuous glucose monitors. For example. CGM can provide substantial glycemic control and benefits for patients with diabetes of different ages. CGMs as an alternative to traditional finger stick tests can track a user’s blood glucose levels every' few minutes. Depending on the device, it may generally predict where the user’s levels will be in the next 10 to 15 minutes and help prevent dangerous fluctuations. Use of CGM can also increase health-related quality’ of life and satisfaction with the treatment regimens and is associated with fewer depressive symptoms. These benefits are closely associated with the frequency of CGM use. Patients who use CGM for the majority ofAttorney Docket No. 3726023.00006 time (generally considered to be 80% or more) have improved glycemic control in the absence of increased hypoglycemia.

[0004] CGMs may make self-management of care decisions easier, but sometimes cause unexpected issues when a patient does not understand how to use her device or struggles with staying on top of her routine. Consistent CGM use remains problematic for many patients. In one aspect, the associations of individual-level social determinants of health (SDoH) and other factors with device use and clinical outcomes are difficult to identify. As a result, actions that tailor patient interventions to improve patient adherence rates for chronic care management can be challenging.

[0005] Accordingly, there is a need for utilizing advanced predictive analytics to analyze comprehensive patient data including transaction data (eg., first party data from supplies history) for CGM, SDoH data, and other clinically-validated information to identify patient behavior and personalize recommendations to improve future health outcomes.Summary

[0006] Among other features, the present disclosure relates to a system deployed within a communication network, the system comprising: a computing device including a non-transitory machine readable storage medium configured to store an application program, and a processor coupled to the non-transitory machine readable storage medium and configured to control a plurality of modules to execute instructions of the application program to obtain and transmit data associated with a patient working with a CGM. The system may also comprise a computing server system configured to obtain the data associated with the patient working with the CGM and SDoH of the patient; and a machine learning (ML) based model, deployed, trained, and configured by the computing server system, to receive the data associated with the patient working with the CGM and the SDoH of the patient, analyze patient behavior and attrition with the CGM over a selected period of time based at least upon the data associated with the patient working with the CGM and the SDoH of the patient, determine, in connection with an observation date, a plurality of predictive models for upcoming 1 to 3 months to identify' a non-adherence event of the patient, determine a risk segment for the patient as a function of the patient behavior and attrition with the CGM over the selected period of time and results of the plurality of predictive models for upcoming 1 to 3 months in connection with the observation date, perform a cohort determination and a persona selection of the patient, and develop tailored intervention plans to engage patients for continuing adherence based on theAttorney Docket No. 3726023.00006 cohort determination and the persona selection of the patient.

[0007] In some embodiments, the risk segment for the patient may be one of: high risk attrition, medium risk attrition, high risk lapsing, medium risk lapsing, low risk, and adherent.

[0008] The computing server system may be further configured to process the data associated with the patient working with the CGM and the SDoH of the patient by removing variables relating to timestamp information, dates, identifier information, Y variables, and missing values, identifying a first portion of the data associated with the patient working with the CGM and the SDoH of the patient as categorical features, and identify ing a second portion of the data associated with the patient working with the CGM and the SDoH of the patient as numerical features.

[0009] Moreover, prior to running the ML based model, the computing server system may be further configured to process the categorical features by removing data that are constant in nature, have high imbalance, contain missing data, have high cardinality, have high categorical association, and relate to data leakage and model tuning.

[0010] In addition, prior to running the ML based model, the computing server system may be further configured to process the numerical features by removing data that contain missing values, have high correlation, have high multi col linearity with other variables, and relate to data leakage and model tuning, and adding selected features from the categorical features as numeric.

[0011] According to one embodiment, the ML based model may be configured to perform the cohort determination of the patient based on the risk segment, an insurance type carried by the patient, and a binary indicator of the patient with respect to the CGM.

[0012] Additionally, the ML based model may be configured to perform the persona selection of the patient in accordance with a plurality of predefined persona groups created based at least on information relating to the patient’s demographic, psychographic and behavioral patterns, channel or engagement preferences, social determinants of health data, and diagnoses and chronic conditions.

[0013] The ML based model may be further configured to perform a Pearson chi-square testing to evaluate the tailored intervention plans in individual cohorts of Medicare, Medicaid, and commercial insurance plan enrollees who had lapsed or been lost to attrition.

[0014] Furthermore, the ML based model may be configured to generate signals representing the tailored intervention plans and transmit the signals to the computing device for display.Attorney Docket No. 3726023.00006

[0015] In yet another embodiment, the ML based model may be configured to determine a plurality of channels for transmitting the signals to the computing device for display, wherein the plurality of channels include a combination of emails, text messages via short message / messaging service (SMS), and outbound interactive voice response.

[0016] According to other aspects, the present disclosure relates to a computing server system deployed within a communication network. An example computing server system may comprise a non-transitory machine readable storage medium storing machine executable instructions; and a processor coupled to the non-transitory machine readable storage medium and configured to control a plurality of modules to execute the machine executable instructions to: obtain data relating to a patient, wherein the data include demographic information, psychographic and behavior patterns, previous recording channel of a CGM and engagement preferences, SDoH, and diabetes diagnoses and specific comorbidities; and deploy and train a machine learning (ML) based model to: receive the data associated with the patient working with the CGM and the SDoH of the patient, analyze patient behavior and attrition with the CGM over a selected period of time based at least upon the data, determine, in connection with an observation date, a plurality of predictive models for upcoming 1 to 3 months to identify a non-adherence event of the patient, determine a risk segment for the patient as a function of the patient behavior and attrition with the CGM over the selected period of time and results of the plurality of predictive models for upcoming 1 to 3 months in connection with the observation date, perform a cohort determination and a persona selection of the patient, and develop tailored intervention plans to engage patients for continuing adherence based on the cohort determination and the persona selection of the patient.

[0017] In certain embodiments, the risk segment for the patient may include one of: high risk attrition, medium risk attrition, high risk lapsing, medium risk lapsing, low risk, and adherent.

[0018] The processor of the computing server system may be further configured to control the plurality of modules to execute the machine executable instructions to process the data by removing variables relating to timestamp information, dates, identifier information, Y variables, and missing values, identifying a first portion of the data as categorical features, and identifying a second portion of the data as numerical features.

[0019] In an embodiment, prior to running the ML based model, the processor may be further configured to control the plurality of modules to execute the machine executable instructions to process the categorical features by removing data that are constant in nature, have highAttorney Docket No. 3726023.00006 imbalance, contain missing data, have high cardinality, have high categorical association, and relate to data leakage and model tuning.

[0020] In another embodiment, prior to running the ML based model, the processor may be further configured to control the plurality of modules to execute the machine executable instructions to process the numerical features by removing data that contain missing values, have high correlation, have high multicollinearity with other variables, and relate to data leakage and model tuning, and adding selected features from the categorical features as numeric.

[0021] The ML based model may be configured to perform the cohort determination of the patient based on the risk segment, an insurance type carried by the patient, and a binary indicator of the patient with respect to the CGM.

[0022] In additional embodiments, the ML based model may be configured to perform the persona selection of the patient in accordance with a plurality of predefined persona groups created based at least on information relating to the patient’s demographic, psychographic and behavioral patterns, channel or engagement preference, social determinants of health data, and diagnoses and chronic conditions.

[0023] According to one implementation, the ML based model may be configured to perform a Pearson chi-square testing to evaluate the tailored intervention plans in individual cohorts of Medicare, Medicaid, and commercial insurance plan enrollees who had lapsed or been lost to attrition.

[0024] In another example, the ML based model may be configured to generate signals representing the tailored intervention plans and transmit the signals to a computing device deployed within the communication network for display.

[0025] Moreover, the ML based model may be configured to determine a plurality of channels for transmitting the signals to the computing device for display, wherein the plurality of channels include a combination of emails, text messages via SMS, and outbound interactive voice response.

[0026] The above simplified summary of example aspects serves to provide a basic understanding of the present disclosure. This summary is not an extensive overview of all contemplated aspects, and is intended to neither identify key or critical elements of all aspects nor delineate the scope of any or all aspects of the present disclosure. Its sole purpose is to present one or more aspects in a simplified form as a prelude to the more detailed descriptionAttorney Docket No. 3726023.00006 of the disclosure that follows. To the accomplishment of the foregoing, one or more aspects of the present disclosure include the features described and exemplary pointed out in the claims.Brief Description of the Drawings

[0027] The accompanying drawings, which are incorporated into and constitute a part of this specification, illustrate one or more example aspects of the present disclosure and. together with the detailed description, serve to explain their principles and implementations.

[0028] Fig. 1 illustrates a systematic diagram of a healthcare management computing system for improving patient adherence rates for chronic care management, according to an exemplary aspect of the present disclosure;

[0029] Fig. 2 illustrates a block diagram of a ML based computing server system, according to an exemplary aspect of the present disclosure;

[0030] Fig. 3 illustrates two computing stages of a ML based computing server system, according to an exemplary aspect of the present disclosure;

[0031] Fig. 4 illustrates an example workflow of a healthcare management computing system for improving patient adherence rates for chronic care management using multiple random forests corresponding to a plurality of ML models, according to an exemplary aspect of the present disclosure;

[0032] Fig. 5 illustrates a diagram of an ensemble based modeling technique, according to an exemplary aspect of the present disclosure;

[0033] Fig. 6 illustrates a plurality of example patient segmentations, according to an exemplary aspect of the present disclosure;

[0034] Fig. 7 illustrates example patient testimonial emails, according to an exemplary aspect of the present disclosure;

[0035] Fig. 8 illustrates an example healthcare provider (HCP) connect email, according to an exemplary7aspect of the present disclosure;

[0036] Fig. 9 illustrates an example text to busy caregivers, according to an exemplary' aspect of the present disclosure;

[0037] Fig. 10 illustrates example clinical leadership emails, according to an exemplary' aspect of the present disclosure;Attorney Docket No. 3726023.00006

[0038] Fig. 11 illustrates a second example HCP connect email, according to an exemplary aspect of the present disclosure; and

[0039] Fig. 12 illustrates a plurality of individual intervention plans and combination plans, according to an exemplary aspect of the present disclosure.Detailed Description

[0040] Various aspects of the present disclosure will be described with reference to the drawings, wherein like reference numerals are used to refer to like elements throughout. In the following description, for purposes of explanation, numerous specific details are set forth in order to promote a thorough understanding of one or more aspects of the present disclosure. It may be evident in some or all instances, however, that any aspects described below can be practiced without adopting the specific design details described below.

[0041] As the diabetes epidemic continues to grow, it is desirable to provide an individual that is dealing with chronic conditions with a predictive, data-driven support that accurately identify risks of non-adherence, including medical device non-adherence, before they become unmanageable and lead to poor outcomes and higher healthcare spending. Data-driven risk stratification has become a core component of chronic disease management in recent years. However, health plans have encountered challenges in identifying when and why certain individuals move up and down the risk ladder. Many plans primarily work with claims data, which can be incomplete from an analytics perspective and offers little insight into why members are straying from their care plans. With limited scope and up to several months of lag time, this claims dataset alone may not be sufficient to get ahead of the exact moment a person starts to show potential issues that are likely to lead to non-adherence with recommended care best practices. Instead of relying too heavily on claims data alone, the present disclosure relates to systems and methods configured to integrate diverse and comprehensive datasets that provide a more comprehensive and current view of patient activities. Example datasets may include but not limited to socioeconomic data to identify non-clinical barriers; pharmacy data to show medication access and adherence patterns; diabetes supply ordering records to indicate therapy adherence; and device data to highlight continual usage of management tools and control of clinical factors, such as blood glucose levels. These datasets used by the present disclosure paint a powerful, holistic, and timely portrait of a patient’s ability to participate in her own care from a clinical and non-clinical perspective, enabling health plans and providers to pinpoint potential trouble spots and dynamically predictAttorney Docket No. 3726023.00006 rising or falling risks of non-adherence.Exemplary' ML Based Computing System and Environment for Improving Patient Adherence Rates for Chronic Care Management

[0042] In one aspect, the present disclosure relates to utilizing ML based predictive analytics to improve patient adherence rates for chronic care management. Fig. 1 illustrates a systematic diagram of a healthcare management computing system 100, deployed within a server-based computing environment and communication network 112, and configured to interact with one or more users 102 via at least one computing device or system 106a, 106b, 106c, ... 106n and a computing server system 114. Leveraging emerging artificial intelligence (Al) and ML to assess individuals predictively and longitudinally, the healthcare management computing system 100 may enable health plans to identify emerging risks of non-adherence and proactively reach out with support for these health plan members to keep them on the right track with their care.

[0043] In certain aspects, a web portal or web-based system may be implemented on the selected computing device or system 106a, 106b, 106c, ... 106n to provide one or more users 102 with personalized access to data, services, and information, via, e.g., a login and accessible through a web browser associated with the computing system 100. For example, the web portal may be accessible via a uniform resource locator (URL) through the web browser with no installation on the selected computing device or system 106a, 106b. 106c. ... 106n required. In one implementation, the web portion may use single codebase for multiple platforms (responsive design for desktops, tablets, and phones) and updates may be instantaneous and do not require any user action for downloading newer versions.

[0044] In yet another embodiment, an application, which may be a mobile or web-based application (e.g., native iOS or Android Apps), may be downloaded and installed on the selected computing device or system 106a, 106b, 106c, ... 106n for instantiating various modules to obtain information and data related to a CGM patient, interact with at least one user 102 of the application, and provide and display ML based predictive analytics results and other information to the user 102. For example, such an application may be used by a CGM patient or a patient having chronic condition(s), at least one caregiver of a patient having chronic condition(s), medical professionals, or other end-users. Automated agents, scripts, playback software, and the like acting on behalf of one or more people may also be users 102. Such a user-facing application of the healthcare management computing system 100 may include a plurality of modules executed and controlled by at least one processor of the hosting computingAttorney Docket No. 3726023.00006 device or system 106a, 106b, 106c, ... 106n for exchanging data with various computing devices and systems deployed within the communicarion network 112, interacting with the user 102 of the application via a plurality of user interface functions and features, and presenting various information to the user 102 via generated audio, video, and tactile signals. Computing device or system 106a, 106b, 106c, ... 106n hosting the mobile or web-based application may be configured to connect, using a suitable communication protocols 110a, 110b and communication network 112, with the computing server system 114, which is further shown in Fig. 2. Here, the communication network 112 may generally include a geographically distributed collection of computing devices or data points interconnected by communication links and segments for transporting signals and data therebetween. Communication protocol(s) 110a and 110b may generally include a set of rules defining how computing devices and networks may interact with each other, such as frame relay, Internet protocol (IP), transmission control protocol (TCP), user datagram protocol (UDP), hypertext transfer protocol (HTTP). It should be appreciated that healthcare management computing system 100 of the present disclosure may use any suitable communication networks, ranging from local area networks (LANs), wide area networks (WANs), cellular networks, to overlay networks and software-defined networks (SDNs), a packet data network (e.g. , the Internet), mobile telephone networks (e.g. , cellular networks, such as 4G or 5G), Plain Old Telephone (POTS) networks, and wireless data networks (e.g. , Institute of Electrical and Electronics Engineers (IEEE) 802.11 family of standards known as Wi-Fi®. WiGig®. IEEE 802.16 family of standards known as WiMax®), IEEE 802.15.4 family of standards, a Long Term Evolution (LTE) family of standards, a Universal Mobile Telecommunications System (UMTS) family of standards, peer-to-peer (P2P) networks, virtual private networks (VPN), Bluetooth, Near Field Communication (NFC), or any other suitable networks.

[0045] The computing server system 114 may be Cloud-based or an on-site server. The term “server” generally refers to a computing device or system, including processing hardware and process space(s), an associated storage medium such as a memory7device or database, and, in some instances, at least one database application as is well known in the art. The computing server system 114 may provide functionalities for any connected devices such as sharing data or provisioning resources among multiple client devices or performing computations for each connected client device. According to one embodiment, within a Cloud-based computing architecture, the computing server system 114 may provide various Cloud computing services using shared resources. Cloud computing may generally include Internet-based computing in which computing resources are dynamically provisioned and allocated to each connectedAttorney Docket No. 3726023.00006 computing device or other devices on-demand, from a collection of resources available via the network or the Cloud. Cloud computing resources may include any type of resource, such as computing, storage, and networking. For instance, resources may include service devices (firewalls, deep packet inspectors, traffic monitors, load balancers, etc.), computing / processing devices (servers, central processing units (CPUs), graphics processing units (GPUs), random access memory, caches, etc.), and storage devices (e.g., network attached storages, storage area network devices, hard disk drives, solid-state devices, etc.). In addition, such resources may be used to support virtual networks, virtual machines, databases, applications, etc. The term “database,” as used herein, may refer to a database (e.g., relational database management system (RDBMS) or structured query' language (SQL) database), or may refer to any other data structure, such as, for example a comma separated values (CSV), tab-separated values (TSV), JavaScript Object Notation (JSON), eXtendible markup language (XML), TeXT (TXT) file, flat file, spreadsheet file, and / or any other widely used or proprietary format. In some embodiments, one or more of the databases or data sources may be implemented using one of relational databases, flat file databases, entity -relationship databases, object-oriented databases, hierarchical databases, network databases. NoSQL databases, and / or record-based databases.

[0046] Cloud computing resources accessible using any suitable communication network (e.g. , Internet) may include a private Cloud, a public Cloud, and / or a hybrid Cloud. Here, a private Cloud may be a Cloud infrastructure operated by an enterprise for use by the enterprise, while a public Cloud may refer to a Cloud infrastructure that provides services and resources over a network for public use. In a hybrid Cloud computing environment which uses a mix of onpremises. private Cloud and third-party, public Cloud services with orchestration between the two platforms, data and applications may move between private and public Clouds for greater flexibility and more deployment options. Some example public Cloud service providers may include Amazon (e.g , Amazon Web Services® (AWS)), IBM (e.g., IBM Cloud), Google (e.g , Google Cloud Platform), and Microsoft (e.g. , Microsoft Azure®). These providers provide Cloud services using computing and storage infrastructures at their respective data centers and access thereto is generally available via the Internet. Some Cloud service providers (e.g . Amazon AWS Direct Connect, Microsoft Azure ExpressRoute) may offer direct connect services and such connections ty pically require users to purchase or lease a private connection to a peering point offered by these Cloud providers.Attorney Docket No. 3726023.00006

[0047] According to one embodiment, referring to Fig. 2, the computing server system 114 may include at least one processor 202 configured to control and execute a plurality of modules to establish a customer analytical record (CAR) for each patient via an analytical record module 208, perform ML based prediction to identify patients at risk of non-adherence in CGM therapy via an ML based predictive model 210, perform adherence-based risk segmentation using a patient risk analyzer 212, and determine tailored intervention plans via an intervention plans determination module 214 to engage patients for continuing adherence. Fig. 3 illustrates two stages of ML based predictive analytics of the present disclosure. In the first stage 302, results may be generated by the ML based predictive model 210 to predict the adherence behavior of a specific patient in the upcoming three months. In the second stage 304, the patient risk analyzer 212 may be configured to predict the segment of the patient based at least upon the adherence and non-adherence determination of the first stage 302. According to one embodiment, six segments (e.g., high risk attrition, medium risk attrition, high risk lapsing, medium risk lapsing, low risk, and adherent) may be defined based on six months of adherence behavior (e.g. , previous three months and future three months) for a particular observation date.

[0048] It should be appreciated that the term “module,'’ “analyzer,” and “model” or other similar terms as used herein may refer to a real-world device, component, or arrangement of components and circuitries implemented using hardware, such as by an application specific integrated circuit (ASIC) or field-programmable gate array (FPGA), for example, or as a combination of hardware and software, such as by a microprocessor or microcontroller system and a set of instructions to implement the module’s functionality, which (while being executed) transform the microprocessor system into a special purpose device. A “module.” “analyzer,” and “model” may also be implemented as a combination of the two, with certain functions facilitated by hardware alone, and other functions facilitated by a combination of hardware and software. Each “module,” “analyzer,"’ and “model"’ may be realized in a variety of suitable configurations, and should not be limited to any example implementation exemplified herein.

[0049] Memory 216, which is coupled to the processor 202, may be configured to store at least a portion of information obtained by the transceiver module 204. In one aspect, memory 216 may be a non-transitory machine readable medium configured to store at least one set of data structures or instructions (e.g. , software) embodying or utilized by at least one of the techniques or functions described herein. It should be appreciated that the term “non-transitory computer or machine readable medium” may include a single medium or multiple media (e.g. , one or more caches) configured to store the at least one instruction. The term “machine readableAttorney Docket No. 3726023.00006 medium’" may include any medium that is capable of storing, encoding, or carrying instructions for execution by all modules of the computing server system 114 and that cause these modules to perform at least one of the techniques of the present disclosure, or that is capable of storing, encoding or carrying data structures used by or associated with such instructions. Non-limiting machine-readable medium examples may include solid-state memories, and optical and magnetic media. Specific examples of machine-readable media may include: non-volatile memory, such as semiconductor memory devices (e.g, Electrically Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM)) and flash memory devices; magnetic disks, such as internal hard disks and removable disks; magneto-optical disks; Random Access Memory (RAM); Solid State Drives (SSD); and CD-ROM and DVD-ROM disks.

[0050] According to some implementations, the transceiver module 204 may be configured by the processor 202 of the computing server system 114 to exchange various information and data with other modules and / or computing devices / sy stems deployed within the communication network 112 and connected with the computing server system 114. The interface 206 may be configured to allow a user to control and access different modules and computing devices connected with the computing server system 114. Various information relating to the control and the processing of data may be presented to the user via the interface 206 which may include any suitable graphical user interface, command line interface, menu-driven user interface, touch user interface, voice user interface, form-based user interface, natural language user interface, and mobile user interface (e.g, graphical and touch -sensitive display features associated with mobile computing devices such as smartphones and tables).

[0051] In accordance with aspects, the analytical record module 208 of the computing server system 114 may establish a CAR to form a patient centric view' of patient behaviors, attitude, characteristics in order to input into the ML based predictive model 210. In connection with each CAR. the computing server system 114 may monitor each patient over time, perform attrition tracking, identify patient characteristics and their impact on adherence, and perform daily cadence tracking. For example, the computing server system 114 may obtain and analyze over 800 features or more than 270 metrics relating to each patient including adherence management and transaction data (e.g, first party data from supplies history) for CGM, SDoH data, and other clinically-validated information. Specifically, the SDoH data generally relate to the conditions in the environments w'here people are bom, live, learn, work, play, worship, and age that affect a wide range of health, functioning, and quality-of-life outcomes and risks.Attorney Docket No. 3726023.00006 Both the 30-day and 90-day models of the present disclosure may utilize selected unique numerical and categorical features based on high correlation, cardinality and statistical significance to the models. These features are unique to the algorithm of the healthcare management computing system 100 and being continuously refined for improved predictions. Example adherence management data of each CAR may include data relating to demographics, insurance, temporal, vitals, provider visits, HCP characteristics, CGM devices, adherence management program / service engagement, medication, shipping information, adherence behavior, and channel preference and history of each patient. Example SDoH data of each CAR may include data relating to demographics, financial health, comorbidities, attitude, and channel preference and history of each patient.

[0052] In some important aspects, the computing server system 114 may be configured to incorporate, host, and train one or more advanced ML based models (e.g., the ML based predictive model 210) to analyze patterns of CGM patients' behaviors in order to proactively (e.g., 3 months in advance) identify patients at risk of non-adherence in CGM therapy and to determine relevant patient intervention plans. The ML based predictive model 210 may be locally deployed or can be accessed via remote application programming interface (API) calls by the computing server system 114. In one embodiment, the computing server system 114 may run the ML based predictive model 210 locally but interact with it through an API, mimicking the structure of remote APIs. This implementation may provide more control over the data and the environment while maintaining the flexibility of API interactions.

[0053] In accordance with some implementations, referring back to Fig. 1, the computing server system 114 (e.g, Cloud-based or an on-site server) of the present disclosure may be configured to connect with various data sources or services 116a, 116b. 116c, ... 116n. For example, the computing server system 114 may be configured to host, train, operate, and / or incorporate any suitable type of Al and / or ML based models (e.g, at least one of 116a, 116b, 116c, ... 116n) via remote API calls to utilize predictive analytics to improve patient adherence rates for chronic care management of the user 102 using comprehensive data points to personalize recommendations and understand patient behavior.

[0054] For example, the computing server system 114 may be configured to track patient characteristics and their impact on adherence rates for chronic care management by at least obtaining and analyzing a wealth of data and deep relationships with patients and their providers, powered by predictive AI / ML models. Adherence may be generally defined as the extent to which a patient's behavior in terms of taking medication, following a diet, modifyingAttorney Docket No. 3726023.00006 habits, or attending clinics coincides with medical or health advice. For example, early warning of future adverse events may be generated by the computing server system 114 of the present disclosure to identify patients declining a re-order of certain CGM supplies. In some aspects, the system and method of the present disclosure predict the risk of non-adherence events 3 months in advance, and continuously track patient behavior and attrition over time.

[0055] As will be described fully below, the present disclosure uses advanced Al and ML analytic techniques to uncover the value of complex adherence patterns of each individual patient. Al has quickly become an essential tool for making sense of rich and varied healthcare datasets, but it must be deployed intentionally to maximize its impact. That means developing algorithms and services that can identify accessible patient data, while also identifying what data are missing in a patient’s longitudinal record.

[0056] For example, the sudden absence of a monthly diabetes supply order or prescription refill, or a sporadic tapering off data reports from a CGM over time, may be major red flags on the adherence front. The ML based predictive model 210 of the present disclosure may be configured to determine when missing data is a sign of an impending non-adherence, which means designing models and corresponding patient outreach and education strategies that support prevention.

[0057] After examining these patterns at scale and over time, the ML based predictive model 210 may be configured to accurately assist health plans to identify clinical and socioeconomic factors that most directly correlate with the adherence gaps in their unique populations, allowing care management teams to move closer and closer to the non-adherence trigger point for individuals. Further, the ML based predictive model 210 may predict likely non-adherence events for patients before they occur.

[0058] In some situations, some patients provide care for children and aging parents while working full-time and may have more limited opportunities to invest in their own care. Offering these patients insights and best practices specific to maintaining therapy under a tight schedule can prove helpful. In other cases, financial uncertainty may be impacting a patient. Providing these patients with education and coaching on tools that allow for flexibility in out-of-pocket costs for medications and / or medical devices so that they can continue therapy without disruption can mean the difference between adherence and non-adherence.

[0059] As will be described fully below, the present disclosure further relates to connecting with patients to prevent and resolve adherence challenges.

[0060] Identifying impending problems may be only half the battle in certain circumstances. Meaningful and individualized outreach may be implemented to members who show signs ofAttorney Docket No. 3726023.00006 non-adherence as soon as possible. Direct engagement and education with members can often uncover the true obstacles, both tangible and emotional, behind non-adherence issues, including underlying issues of trust in the health system that may stem from personal or community experiences. These interactions with extended care teams may shift that narrative for individuals and become an opportunity for plans to provide compassionate, actionable problem-solving for members that help build relationships and prevent future issues.

[0061] Information gathered during these outreach interactions can be structured and input into analytics efforts to enrich future insights and enable health plans to become even more predictive, personalized, and prepared to support their members with community-based resources, tailored diabetes education, and specialized training on how to best use their devices and adhere to a recommended care regimen.

[0062] The ML based predictive model 210 used by the present disclosure may be configured to identify non-adherence issues in people living with diabetes before it becomes a full blown, costly problem. By analyzing holistic datasets and member care patterns, the ML based predictive model 210 identifies the underlying challenges facing patients and empowers health plans to address these issues earlier, while fostering meaningful outreach activities that surround people living with diabetes with the support and guidance they need to thrive.

[0063] As an evidence-based approach, the healthcare management computing system 100 of the present disclosure applies predictive analytics to improve risk stratification and create patient cohorts for proactive intervention and improved adherence.

[0064] For example, based on a random forest model or classifier or any suitable ML algorithms, the ML based predictive model 210 may be parameterized with a plurality of selected input features (e.g, at least 108 input features based on the CAR) based on automated feature engineering with input data pipelines, prediction pipelines, retraining pipelines and model monitoring capabilities constantly evaluating 900+ features to arrive at the suitable input used to predict 30 days for Month 1, Month 2, Month 3 for each patient. For example, the random forest classifier of the present disclosure may be employed for extracting features, training, and forecasting sample data using multiple decision-making trees. In an embodiment, the ML based predictive model 210 may include a 30-day model configured to identify a group of patients' adherence in the past 30 days and separately predict those patients’ adherence in the next 30 days (Model 1), next 60 days (Model 2), and next 90 days (Model 3). In another embodiment, the ML based predictive model 210 may include a 90-day model configured to identify' a group of patients’ shipment records and partition them into the past 30 days (ModelAttorney Docket No. 3726023.00006 1), past 60 days (Model 2), past 90 days (Model 3), and beyond 90 days (Model 4) and predicts the month of shipment for a patient in the next 90 days. Lookback period for both the 30-day model and the 90-day model may be up to 12 months. Predictions may be available every first week of a month in some implementations. Model training cadence may not be fixed. These models may be trained by e.g., the computing server system 114 in response to detecting a drift in model performance.

[0065] In some embodiments, a combination of binary and multi-class classification models may be used to map inputs to the ML based predictive model 210 to a plurality selected class labels after obtaining unique datasets from various proprietary sources and external data sources relating to patient features and observations. Fig. 4 illustrates an example workflow of a healthcare management computing system for improving patient adherence rates for chronic care management using multiple random forests corresponding to a plurality of ML models. In one aspect. 7 different random forests 1-7 corresponding to a 30-day Model 1 402 which may be configured to predict patient adherence in the next 30 days, a 30-day Model 2 404 which may be configured to predict patient adherence in the next 60 days, a 30-day Model 3 406 which may be configured to predict patient adherence in the next 90 days, a 90-day Model 4 408 which tracks patient last shipment of CGM devices and supplies in the past 30 days and predicts the month of shipment for the patient in the next 90 days, a 90-day Model 5410 which tracks patient last shipment of CGM devices and supplies in the past 60 days and predicts the month of shipment for the patient in the next 90 days, a 90-day Model 6 412 which tracks patient last shipment of CGM devices and supplies in the past 90 days and predicts the month of shipment for the patient in the next 90 days, and a 90-day Model 7414 which tracks patient last shipment of CGM devices and supplies over 90 days and predicts the month of shipment for the patient in the next 90 days, respectively.

[0066] Next, using the combination of binary and multi-class classification models, the computing server system 114 may use bagging methodology and random forest classifiers to reduce both variance and avoid overfitting issues typically occurred in classification models. For example, the computing server system 114 may use bootstrap aggregation, a ML technique that improves the accuracy and stability of ML algorithms. As shown in Fig. 5, multiple decision trees (e.g., 30 decision trees according to one implementation) may be used to perform data classification, in accordance with aspects of the present disclosure.

[0067] Randomness may be used to enhance the ML based predictive model's robustness, enable efficient optimization, reduce overfitting, and allow for more realistic data variation. InAttorney Docket No. 3726023.00006 one aspect, randomness may be achieved by the computing server system 114 in two ways: for each decision tree, rows are randomly selected; and for each custom split, predictive inputs are randomly selected. This method reduces both variance and avoids overfitting in classification models, thereby generating more precise likelihood of patient adherence based on unique patient features input into the computing server system 114.

[0068] According to some embodiments, for each decision tree, the ML based predictive model 210 may perform bootstrap sampling where anew subset of the original dataset may be created by randomly sampling rows (data points) with replacement from the dataset. Since sampling is done with replacement, some rows may be selected multiple times in the subset, while others may not be selected at all. As a result, each decision tree may have a slightly different view of the dataset. In one aspect, the ML based predictive model 210 may be configured to make random row selection to achieve randomness. That is, by training each tree on a different subset of the data, the trees may become diverse and make different errors, which reduces the risk of overfitting. When the predictions of multiple diverse trees are aggregated, the ML based predictive model 210 may achieve better generalization and more accurate predictions than a single decision tree trained on the entire dataset. Bagging refers to the process of creating multiple models from different subsets of the data and then averaging their predictions (in the case of regression) or taking a majority vote (in the case of classification). As shown in Fig. 5, the ML based predictive model 210 may generate prediction results 502 using a majority voting technique 504.

[0069] Moreover, when building a decision tree, each node (or split) may evaluate all available features to find the best one to split on. However, in a random forest, only a random subset of features is considered at each split, rather than all features. For example, if a dataset has 10 features, instead of evaluating all 10 at each split, the ML based predictive model 210 may evaluate a randomly selected subset of, e.g., 3 features, to achieve random selection of features at each split. This random feature selection introduces diversity among the decision trees, as each tree will likely use a different subset of features for its splits. By forcing trees to split on different features, random forests of the present disclosure may become less sensitive to any particular feature, making the ensemble less likely to overfit to the training data. Further, since each tree may consider a random selection of features at each split, the decision trees of the present disclosure may develop in diverse ways. This model independence among trees may improve the ML based predictive model 210’s ability to generalize to new data. If certain features are highly predictive, standard decision trees may tend to select those features at everyAttorney Docket No. 3726023.00006 split. By limiting the choice of features at each split, the ML based predictive model 210 may prevent any single feature from dominating the structure of the trees, which can help improve the ensemble’s overall accuracy and robustness.

[0070] In some implementations, Gini impurity may be used to quantify the likelihood of an incorrect classification of a randomly chosen element from the dataset if it were labeled according to the class distribution in that node. If all elements in a node belong to the same class, the Gini impurity is 0 (pure node). If the elements are split evenly across classes, the impurity is higher. Gini impurity may be used to guide the decision tree algorithm of the present disclosure to decide where to split the data by evaluating how well a potential split will separate different classes.

[0071] As shown in Fig. 5, using an ensemble-based voting classifier (e.g, majority voting technique 504), the ML based predictive model 210 may combine the outputs of e.g., the 30 individual trees to reach a single result 502. In one implementation, 30 models of patient behavior types are accounted in predicting patient adherence. For a node with k classes, the Gini impurity G is calculated as the following, where p is the proportion of instances belonging to class i in the node. This formula essentially calculates the probability of a misclassification.

[0072] When building a decision tree, the algorithm evaluates potential splits by calculating the Gini impurity for each possible split. The goal is to find the split that results in the lowest Gini impurity in the resulting child nodes, which indicates the data is more homogeneously classified within each node. For example, if a node has a mix of classes, a split that separates the classes more effectively will reduce the Gini impurity in the resulting nodes. It should be appreciated that other metrics may be used to evaluate node purity, such as entropy. Lower Gini impurity or entropy indicates purer nodes indicating better separation of classes.

[0073] In yet another aspect, the ML based predictive model 210 may be configured to use distributed ML algorithms and utilities, such as Spark based MLlib, that are optimized to run in parallel across a cluster of computers. This makes it especially useful for processing large datasets to accommodate recent patient behavior towards adherence. For example, patient-month combination over last 24 months results in training data close to 5 million rows. The ML based predictive model 210 may utilize distributed ML algorithms to perform data processing and feature engineering including feature extraction, feature selection,Attorney Docket No. 3726023.00006 normalization, and standardization. In an embodiment, the ML based predictive model 210 may use an API to build ML workflows that involve multiple steps (e g., data transformations, feature engineering, and model training). This makes it easy to create reusable workflows and simplifies the training and tuning of models. The ML based predictive model 210 may also be configured to evaluate model performance and tuning hyperparameters. As disclosed above, the ML based predictive model 210 may use Spark based MLlib and random forest algorithm for classification, thereby supporting both binary and multiclass labels, as well as both continuous and categorical features. In one embodiment, the ML based predictive model 210 may set a seed which is an initial value used by a random number generator to control the randomization processes within the algorithm. Setting a seed may allow the random forest model of the present disclosure to produce the same results each time it is run, given the same input data. The train-test split ratio may be set to 80 / 20 (80% training, 20% test) or 70 / 30 (70% training, 30% test), while the exact ratio may vary depending on the dataset size, model complexity, and the need for a larger or smaller training set.

[0074] In one aspect, input features to the analytical record module 208 may be selected before the ML based predictive model 210 runs. It should be appreciated that the input features to the computing server system 114 may be obtained from various proprietary and external data sources continuously and refined over time. According to one implementation, an initial set of input features to the analytical record module 208 may include over 800 features. A filtering process may be performed by the computing server system 114 to remove certain variables or features (e.g, timestamp information, dates, identifier information, Y variables, and missing values) from the CAR input features. Categorical and numeric feature selections may be carried out. Here, the categorical features, variables or data may inherently embed prior medical information during its process of categorization, whereas the numerical features or data may be flexible for accurate measurements and reading. In an embodiment, 73 out of 518 categorical features may be included in the ML based predictive model 210 after removing 34 features due to constant in nature (e.g., only one value for all the outputs or target values), 85 features due to high imbalance, 1 feature due to missing data, 49 features due to high cardinality (insurance number, city, zip code, etc.), 222 features due to high categorical association using Cramer' s V, 36 features due to data leakage, and 19 features in model tuning ( 1 month indicator added). Additionally, 33 out of 184 numeric features may be included in the ML based predictive model 204 after removing 3 features because of missing values, 129 features due to high correlation, 12 features due to high multicollinearity with other variables, 2 features dueAttorney Docket No. 3726023.00006 to data leakage (number of consents), 6 features in model tuning, and adding 1 feature from categorical feature as numeric.

[0075] In one embodiment, statistical analysis may be performed on the numeric features based on Spearman’s correlation to assess how well the relationship between two variables can be described using a monotonic function (whether linear or not). Specifically, the Spearman correlation between two variables will be high when observations have a similar (or identical for a correlation of 1) rank between the two variables, and low when observations have a dissimilar (or fully opposed for a correlation of-1) rank between the two variables. Further, in the context of ML, multi-collinearity refers to a high correlation between predictor variables in a regression model, and data leakage refers to a problem where information from outside the training dataset is used to create the model which can lead to overly optimistic performance estimates during training and validation, as the model has access to extra information. As a result, a total of 106 features may be included in the in the ML based predictive model 210 where 33 features have been identified as numeric and 73 features are identified as categorical.

[0076] Among all of the 106 input features to the analytical record module 208, a number of features may carry more weights in predicting a patient’s adherence in next 30 days with high accuracy. For example, a plurality of feature domains may be identified by the ML based predictive model 210, including insurance related information, patient’s adherence behavior, patient’s communication channel preference and history, adherence management program / service engagement, CGM device related information, and marketing communication. Each of these feature domains may have one or more subdomains to correspond to patient specific data and behavior patterns.

[0077] According to one implementation, top 15 features of the 30 days model of the present disclosure may include the following. The feature domain of a patient’s adherence behavior may include but not limited to adherence in the last 3 months, average days between shipments in last 6 months, and total tenure of patient with adherence management program / service engagement. The feature domain of a patient’s communication channel preference and history may include but not limited to the number of channel emails sent in last 6 months, the number of outbound interactive voice response (IVR) calls in last 1 month, total call duration in minutes in the past 12 months, and total agent call duration in minutes in the past 12 months. The feature domain of insurance related information may include but not limited to reimbursement from insurance for an insured, the total expected price for the last shipment, the number of times a patient logged in portal in last 12 months, and insurance parent plan. The featureAttorney Docket No. 3726023.00006 domain of adherence management program / service engagement may include but not limited to patient engagement - regular patient, and engagement billing hold. The feature domain of CGM device related information may include but not limited to the number of days since a patient is using the same brand of CGM device. The feature domain of marketing communication may include but not limited to marketing Genesys calls in last 1 month.

[0078] In one implementation, top 10 features of the 90-day model used for identifying a patient's adherence in the last month from an observation date may include the following: time on current CGM device and current brand; patient engagement - regular patient; average days between shipments during prior 12 months; insurance most recent expected price; number of outbound calls through agent in the past; number of marketo emails sent in the past 6 months; CGM:: current device; adherence management program / service engagement billing hold; number of adherences in the past 12 months; and insurance parent plan - missing.

[0079] In one implementation, top 10 features of the 90-day model for identifying a patient’s adherence in the last two months from an observation date may include the following: CGM Device - Current Device; number of adherences in the past 12 months; number of outbound calls through agent in the past; time on current CGM device and current brand; expected price (shipping summary); adherence management program / service engagement billing hold; patient engagement - regular patient; number of portal visits in the past 12 months; average days between shipments; and number of marketo emails sent in the past.

[0080] According to another implementation, top 10 features of the 90-day model for identify ing a patient’s adherence in the last three months from an observation date may include the following: number of outbound calls through agent in the past; number of marketo emails sent in the past; CGM::current device; adherence management program / service engagement billing hold; number of portal visits in the past 12 months; insurance most recent expected pnce; insurance parent plan - Humana; average days between shipments during prior 12 months; CGM: Current Device; and insurance parent plan - missing.

[0081] In yet another implementation, top 10 features of the 90-day model for predicating a patient’s failure to reorder from an observation date may include the following: number of adherences in the past 12 months; patient engagement - regular patient; number of marketo emails sent in the past; insurance parent plan - missing; number of portal visits in the past 12 months; num_features_insurance_num_primary_insurance_name-change_L12; time on current CGM device and current brand; average days between shipments during prior 12Attorney Docket No. 3726023.00006 months; adherence management program / service engagement billing hold; and insurance most recent expected price.

[0082] Thereafter, the computing server system 114 may identify multiple adherence-based risk segments and perform patient segmentation via the patient risk analyzer 212 accordingly. As shown in Fig. 6, a first group of identified segments may include patients that have been determined to have high or medium risk attrition. A second group of segments may include patients that have been determined to have high or medium risk lapsing. A third segment may include adherent patients. The risk segment a patient falls into may depend on the patient's adherence in the last 3 months and the model’s prediction of the patient’s adherence in the next 3 months. For example, a patient who was adherent for 2 months in the past 3 months and for whom the ML based predictive model 210 predicts will be adherent for the next 3 months falls into ‘"medium risk of lapsing” category’, as identified by the patient risk analyzer 212.

[0083] In connection with each patient’s risk segment identification, the intervention plans determination module 214 of the computing server system 114 may be configured to develop a continuous and comprehensive intervention program. Example different interventions, alone and in combination, may be utilized such as instructions for patients (emails and programmed learning); increased communication and counseling (compliance therapy; automated telephone, computer-assisted patient monitoring and counseling; manual telephone follow-up; family intervention); any suitable approaches to increase the convenience of care (provision at the worksite, simplified dosing); involving patients more in their care through self-monitoring; reminders (tailoring the regimen to daily habits; special reminder pill packaging; dosedispensing units of medication and medication charts; appointment and prescription refill reminders); and reinforcement or rewards for both improved adherence and treatment response (e.g., reduced frequency of visits and partial payment for blood pressure monitoring equipment).A Persona-Based Digital Intervention Program

[0084] Currently, diabetes therapies have improved markedly in the past several decades, with options that now include long, intermediate, and short acting insulin formulations, insulin analogs with improved pharmacokinetic profiles, and an array of oral medications such as sulfonylureas and biguanides (e.g., metformin). These advances have greatly improved glycemic control, mitigated treatment complications, and improved the quality of hfe for diabetes patients. The utility of these therapies is significantly enhanced by7the concurrent useAttorney Docket No. 3726023.00006 of CGM devices, which offer critical real-time monitoring of glucose levels and fluctuations throughout the day, allowing for more timely decisions about treatment than would be possible with traditional intermittent glucose monitoring techniques. CGM helps patients achieve tighter glycemic control without the fear of hypoglycemia and is associated with reduced medical costs and healthcare utilization. CGM can also reduce the burden of glucose monitoring, as it obviates the need for the all-to-familiar discomfort of frequent finger sticks. However, CGM can only improve outcomes if patients use it, and previous studies have shown that nonadherence to CGM is associated with adverse health outcomes and increased costs due to less consistent glycemic control. Moreover, although CGM use can reduce costs, it has been estimated that approximately one-quarter of expenditures for CGM supplies are wasted during the first year after an initial order of supplies because they are paid for but not used. As described above, the ML based predictive models of the present disclosure can successfully predict CGM reordering behavior in patients with diabetes based on data obtained from a large chronic care management company. As a result, these models reliably identify, several months in advance, patients at risk of nonadherence with CGM use and the specific clinical, psychological, administrative, and socioeconomic variables associated with that increased risk. Based on these determinations, the computing server system 114 may develop, implement, and evaluate patient interventions tailored to the identified individual risk categories.

[0085] In an example study, data on CGM ordering practices and insurance type may be obtained (e.g., via the transceiver module 204 of the computing server system 114) for adult patients (aged >18 years) with diabetes who ordered CGM supplies through a large chronic care management company (e.g.. CCS Medical, Dallas, TX). Additionally, a third-party provider of consumer behavior data supplied additional inputs to help identify attributes and characteristics associated with the risk of non-adherence.

[0086] Based on their ordering behaviors, patients may be categorized as adherent, lapsed, or likely to attrite, as described above. Adherence is defined as a 100% reordering rate. Lapsed patients may miss at least one scheduled reorder during the previous 3-month period, and continue to miss one or more in the next 3 months. Attrition patients missing 3 months of shipment / adherence and will predict to miss one or more months in next 3 months. Patients may further be stratified by insurance ty pe (Medicare, Medicaid, or commercial). That is, adherent, lapsed, and attrition patients may be further subdivided into Medicare enrollees, Medicaid enrollees, or commercial plan members, with a total of 9 patient cohorts.Attorney Docket No. 3726023.00006

[0087] Distinct profiles, or “personas,” may be determined using proprietary data on patient behaviors and licensed data from a third-party supplier of consumer data. Determinants of patient personas are composites of variables in the following example categories: demographics (e.g. , age, gender, race / ethnicity, marital status, household size, etc.) psychographic and behavioral patterns; previous reordering channel / engagement preferences (e.g. , email, telephone, mail, in-person); social determinants of health (e.g. , economic status, education level, location); and diabetes diagnoses and specific comorbidities (e.g., obesity, arthritis, cardiac and kidney disease). The resulting personas may be tested as part of a scalable risk identification and content validation process. The personas of the present disclosure may change, evolve and editable. In one embodiment, the computing server system 114 may be configured to continuously identify characteristics of each persona and fit patients with the personas on a regular basis (e.g. , every 6 months to 1 year).

[0088] Interventions may be developed by the intervention plans determination module 214 of the computing server system 114 to encourage patients categorized into each distinct persona to either return to full adherence (lapsed patients) or reengage with CGM device use (patients lost to attrition). Patient cohort determination and persona selections may be carried out by the ML based predictive model 210 of the computing server system 114 to identify the right patient.

[0089] In accordance with further aspects, the present disclosure defines a plurality of cohorts based at least upon risk segment of a patient (e.g. , adherent, lapsing, attrition), insurance type carried by the patient (e.g., Medicare / Medicare replacement plans, commercial or Medicaid), and a binary indicator (e.g. , new or not new) of the patient with respect to CGM. Example cohorts may include but not limited to: (Attrition. Medicare, not new to CGM); (Lapsing, Medicare, Not New to CGM); (Adherent, Medicare, Not New to CGM); (Attrition, Medicare, New to CGM); (Lapsing, Medicare, New to CGM); (Adherent, Medicare, New' to CGM); (Attrition, Medicaid, Not New' to CGM); (Lapsing, Medicaid, Not New to CGM); (Adherent, Medicaid, Not New' to CGM); (Attrition, Medicaid, New to CGM); (Lapsing, Medicaid, New to CGM); (Adherent, Medicaid. New to CGM); (Attrition. Commercial. Not New to CGM); (Lapsing, Commercial, Not New to CGM); (Adherent, Commercial, Not New to CGM); (Attrition, Commercial, New to CGM); (Lapsing, Commercial, New to CGM); and (Adherent, Commercial, New7to CGM). In one implementation, attrition may refer to patients who has not reordered in the last 3 months and will most likely not recorder in the next 3 months. Lapsing may refer to patient who has not reordered in the last 3 months and is likely to recorder in the next 3 months.Attorney Docket No. 3726023.00006

[0090] Messages targeting at-risk patients may be generated based on the risk factors identified for each cohort (lapsed or attrition). Messages may be disseminated through channels patients had utilized in previous healthcare-based interactions. Example channels may include emails, vendor websites, supplier patient portals, telephone communications, and in-person interactions. Contacts with patients may be made at specific times in the patient’s ordering cycle and at times of day when patients have been known to engage their preferred channels. That is. timing optimization for contacts with patients - new patient vs. established patients, or where in recorder journey, or time of day - may be conducted by the computing server system 114 to improve patient adherence and retention.

[0091] In various embodiments, interventions may be tested in patients who had ordered CGM supplies previously but missed orders (lapsed) or had been lost to attrition. CGM reordering after an intervention has been measured as a proxy for a return to CGM adherence. Pearson chi-square testing may be used to evaluate the relative effectiveness of interventions in individual cohorts of Medicare, Medicaid, and commercial insurance plan enrollees who had lapsed or been lost to attrition.

[0092] In one example study, tailored interventions have been developed for three prominent personas identified by the ML based predictive model 210 of the present disclosure: health information seekers who trust patient peers or patient advocacy groups; health information seekers who trust the information provided by health advisors; busy suburban and urban dwellers who care for children (age <18 years) or 1 or more aging family members.

[0093] Patients who seek healthcare information and trust their patient peers or patient advocacy groups may receive a series of emails with patient testimonials or directing them to short, filmed patient testimonials on a durable medical equipment (DME) supplier website, as shown in Fig. 7. The emails may include links to the DME supplier portal with a “Call to Action” encouraging patients to reorder scheduled supplies or register on the DME supplier portal to enable reordering.

[0094] Patients who seek health information and trust health advisors may receive two emails each, as shown in Fig. 8. Depending on the patient’s preferred engagement channel, the emails either directed them to a “Coffee Chat” webinar with a pharmacist or inform them that they would receive a short call from a certified diabetes care education specialist.

[0095] Busy urban caregivers of children or aging family members may receive three text messages reminding them not to neglect their own care as they care for others, as shown in Fig.Attorney Docket No. 3726023.00006 9. The texts may include links to the DME supplier website. These text messages have been sent to 1.133 subjects in Lapsed or Attrition Medicare, Medicaid, and commercially insured cohorts, with 364 patients (32.1%) reordering CGM supplies.

[0096] A systematic use of ML techniques may identify risk factors for patient’s nonadherence with scheduled CGM reordering and develop highly personalized interventions to encourage adherence in patients falling into certain predefined risk categories. Among other features, the ML based predictive model 210 of the present disclosure identifies key predictors of adherence including the number of prescribed medications, outpatient visits, laboratory values, and another critical element in diabetes care, CGM usage which may provide real-time glucose data and trends, enabling better-informed treatment decisions and improved health outcomes. Data from studies have shown that consistent use of CGM is closely linked to significant reductions in A2C levels, medical expenses, and burden on the healthcare system. The use of machine learning to identify adherence patterns in patients using CGM and help develop highly targeted interventions to improve adherence with CGM reordering may help leverage the full potential of CGM to improve diabetes care. Moreover, from the payer’s perspective, improving adherence is also essential for ensuring that the money spent does not go to waste and, as a result, that potential cost savings associated with CGM use are realized. The cost of CGM is largely paid upfront with the initial order. If patients order but lapse or discontinue, that upfront expenditure is wasted.

[0097] Further, in addition to a number of factors associated with CGM adherence including age, percentage of time in glucose target, the perceived necessity of CGM, body mass index (BMI), and sex, the present disclosure shows that ordering CGM supplies through a DME supplier may also improve adherence with CGM reordering and glucose control while reducing healthcare costs. As discussed above, the present disclosure identifies lifestyle and personality characteristics as determinants of adherence, allowing for the creation of interventions to address issues of nonadherence posed by these characteristics.

[0098] In some aspects, the interv ention plans determination module 214 may generate and distribute a variety of campaign-based interventions with a variety of channels including but not limited to emails, short message / messaging service (SMS), outbound IVR, etc. Content for campaigns may be generated in an individual way based on intervention theme but specifically and automatically generated based on the individual patient profile. Further, content may be generated to include either texts for emails or SMS or may be in the form of video and made available to a specific patient. According to one embodiment, referring backAttorney Docket No. 3726023.00006 to Fig. 1, contents of the determined intervention plans may be displayed to the user 102 via an interface of the at least one computing device or system 106a, 106b, 106c. ... 106n. Further, large language models (LLMs) (e.g., at least one of 116a, 116b, 116c, ... 116n of Fig. 1) may be prompted by the computing system 100 of the present disclosure for specific content based on targeted patients who can access the computing system 100. Specific messaging may be distributed via any suitable platform systems.

[0099] In one embodiment, at least one retrieval augmented generation (RAG) approach may be used to improve the efficacy of the LLM applications by leveraging unique patient adherence data set to develop specific content based on targeted patients.

[0100] In certain aspects, the present disclosure may utilize intelligent designs for patient engagement. Continuous identification of at-risk patient segments may be implemented to refine the engagement models that will proactively avoid chum and enhance loyalty of the patients. For example, the ML based predictive model 210 may be used to refine analytics such that target segmentation of patients via the patient risk analyzer 212 may be based on patient adherence behavior. Among other features, the ML based predictive model 210 of the present disclosure may identify and segment target populations and identify engagement opportunities based on the current state of each patient at a personal and individual level. In analyzing the ML based predictive model 210's outputs, the computing server system 114 may also hypothesize the root cause of chum by each segment. In one aspect, proactive intervention plans may be created based on prioritized targeted segment. For example, the ML based predictive model 210 may cany out the implementation of the intervention plans, measure and refine the interventions in accordance with a test-and-leam approach in patient experience.

[0101] Furthermore, the computing server system 114 may identify' key differentiator in managing patient attrition and lapsing. For example, a learning agenda may be used to identify patient cohort and intervention plans and use a champion-challenger approach to identity' new intervention plans. In one embodiment, an intervention backlog may be generated and interest threshold may be determined. Thereafter, for risk management, experiments may be carried on a small percentage of customer base. The computing server system 114 may create next set of items on the intervention backlog and the learning agenda to minimize risk and maximize value. In establishing credibility threshold and proof of value, successful interventions implementations in the ML based predictive model 210 may be recognized and unsuccessful interventions may be discarded. In one preferred embodiment, such a test-and-leam approachAttorney Docket No. 3726023.00006 in patient experience may be conducted by the computing server system 114 quarterly or in accordance with any selected interval of time.

[0102] In one aspect, testing model efficacy of the present disclosure involves determining if various targeted patient outreach emails generate a significant change in ordering behavior. In an example clinical leadership intervention outreach email, as illustrated in Fig. 10, selected health representatives that have access to the computing system 100 of the present disclosure may provide communications and counseling to specific patients. Moreover, in an example HCP connection intervention outreach email, as illustrated in Fig. 11, patient specific HCP names may be included in communication to encourage reorder of supplies for CGM therapy.

[0103] As described above, a plurality of personas may be implemented for determining personalized recommendations based at least upon demographic information, psychographic and behavioral characteristics, channel / engagement preferences, SDoH data, and diagnosis and chronic conditions. Example personas may include but not limited to patients who are proactive in health and wellness, health information seekers who have trust in patient peers, health information seekers who have trust in health care advisors, patients living in urban areas and acting as caretakers, patients new to CGM (1 shipment), patients on older devices (e.g, Medtronic CGM devices), attrition Humana patients who have at least 5 shipments that are proactive in health and wellness, patients over the age of 50 and not consenting via portal of any adherence management program / service engagement, patients whose annual income is less than $50K indicating financial constraints, patients having chronic conditions (e g, arthritis), patients having risk of loneliness, patients likely to have food insecurity, patients filled a prescription via a supermarket or pharmacy and filled a prescription via a mail order; and patients showing likelihood of not having access to transportation.

[0104] Test and control groups for attrition and lapsing patients may be analyzed. According to one embodiment, referring to Fig. 12. 3 individual intervention plans and 3 distinct combination plans may be identified. Specifically, a first example intervention program 1202 for test groups may be related to clinical leadership and emails may be generated and transmitted by the computing system of the present disclosure to each patient including but not limited to introduction emails, newsletters and diabetes day. A second example intervention program 1204 may include HCP connection and emails may be generated and transmitted by the computing system to each patient including but not limited to HCP emails and HCP reminder emails. A third example intervention program 1206 may include educational access where Webinars regarding the management of diabetes may be recommended to each patient.Attorney Docket No. 3726023.00006 In one embodiment, a revised Webinar messaging may be generated to highlight the Q&A portion. Further, emails may be generated and transmitted by the computing system to each patient including but not limited to Webinar invites and Webinar reminders. In one aspect, a tailored patient intervention plan may include a combination of the first, second and third intervention programs. For example, Fig. 12 shows an example combination plan may include clinical leadership andHCP connection 1208, orHCP connection and educational access 1210, or clinical leadership and educational access 1212.

[0105] Each patient’s behaviors and responses to a tailored intervention plan may be tracked by the computing server system 114, as one of the input features to the analytical record module 208, and utilized by the ML based predictive model 210.

[0106] In additional embodiments, the healthcare management computing system 100 may generate Al based content for personalized interventions. Such Al based content may be available in private cloud or analytics platform. Upon receiving patient persona content and motivational factors, the computing server system 114 may extract texts from the inputs and prompt large language models (e.g., open ai hosted by at least one of 116a, 116b, 116c, ...116n of Fig. 1) with appropriate content. AI analysis may be returned to the computing server system 114 for creating patient outreach emails which are personalized and based on specific customer information and generated from services. The output from the computing server system 114 may be communicated through enterprise communications layer of the system.

[0107] It should be appreciated that the ML based predictive model 210 of the present disclosure may not only be applied to CGM device products but also expand to other diabetes related products like blood glucose monitors (BGMs), strips, insulin, lancets or other relevant products. In yet another embodiment, the present disclosure may expand to include diabetes BGM users that could be good candidates for CGM use.

[0108] The present disclosure may be used to manage chronic care conditions with recurring consumables similar to CGM. These may include: sleep with CPAP devices and consumables like masks, filters, hoses, respiratory; COPD with inhalers, nebulizers, and oxygen therapy; dialysis patients with catheters, tubing, etc.-, weight management with related medications; and osteoarthritis or orthopedic pain management with braces, splints, relief patches, heat pads, etc.

[0109] Unless specifically stated otherwise as apparent from the foregoing disclosure, it is appreciated that, throughout the present disclosure, discussions using terms such as “processing,” “computing.” “calculating,” “determining,” “displaying,” or the like, refer to theAttorney Docket No. 3726023.00006 action and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (electronic) quantities within the computer system’s registers and memories into other data similarly represented as physical quantities within the computer system memories or registers or other such information storage, transmission or display devices.

[0110] One or more components may be referred to herein as "configured to," "configurable to," " operable / operative to," "adapted / adaptable," "able to," "conformable / conformed to." etc. Those skilled in the art will recognize that "configured to" can generally encompass activestate components and / or inactive-state components and / or standby-state components, unless context requires otherwise.[OHl] Those skilled in the art will recognize that, in general, terms used herein, and especially in the appended claims (e.g., bodies of the appended claims) are generally intended as "open" terms (e.g., the term "including" should be interpreted as "including but not limited to," the term "having" should be interpreted as "having at least," the term "includes" should be interpreted as "includes but is not limited to," etc ). It will be further understood by those within the art that if a specific number of an introduced claim recitation is intended, such an intent will be explicitly recited in the claim, and in the absence of such recitation no such intent is present. For example, as an aid to understanding, the following appended claims may contain usage of the introductory phrases "at least one" and "one or more" to introduce claim recitations. However, the use of such phrases should not be construed to imply that the introduction of a claim recitation by the indefinite articles "a" or "an" limits any particular claim containing such introduced claim recitation to claims containing only one such recitation, even when the same claim includes the introductory phrases "one or more" or "at least one" and indefinite articles such as "a" or "an" (e.g. , "a" and / or "an" should typically be interpreted to mean "at least one" or "one or more"); the same holds true for the use of definite articles used to introduce claim recitations.

[0112] In addition, even if a specific number of an introduced claim recitation is explicitly recited, those skilled in the art will recognize that such recitation should typically be interpreted to mean at least the recited number (e.g., the bare recitation of "two recitations." without other modifiers, typically means at least two recitations, or two or more recitations). Furthermore, in those instances where a convention analogous to "at least one of A, B, and C, etc." is used, in general such a construction is intended in the sense one having skill in the art would understand the convention (e.g.. "a system having at least one of A, B. and C" would include but not be limited to systems that have A alone, B alone, C alone, A and B together, A and CAttorney Docket No. 3726023.00006 together, B and C together, and / or A, B, and C together, etc.). In those instances where a convention analogous to "at least one of A, B, or C, etc." is used, in general such a construction is intended in the sense one having skill in the art would understand the convention (e.g., "a system having at least one of A, B, or C" would include but not be limited to systems that have A alone, B alone, C alone, A and B together, A and C together, B and C together, and / or A, B, and C together, etc.). It will be further understood by those within the art that typically a disjunctive word and / or phrase presenting two or more alternative terms, whether in the description, claims, or drawings, should be understood to contemplate the possibilities of including one of the terms, either of the terms, or both terms unless context dictates otherwise. For example, the phrase "A or B" will be ty pically understood to include the possibilities of "A" or "B" or "A and B."

[0113] With respect to the appended claims, those skilled in the art will appreciate that recited operations therein may generally be performed in any order. Also, although various operational flow diagrams are presented in a sequence(s), it should be understood that the various operations may be performed in other orders than those which are illustrated, or may be performed concurrently. Examples of such alternate orderings may include overlapping, interleaved, interrupted, reordered, incremental, preparatory, supplemental, simultaneous, reverse, or other variant orderings, unless context dictates otherwise. Furthermore, terms like "responsive to," "related to," or other past-tense adjectives are generally not intended to exclude such variants, unless context dictates otherwise.

[0114] It is worthy to note that any reference to "one aspect," "an aspect," "an exemplification," "one exemplification," and the like means that a particular feature, structure, or characteristic described in connection with the aspect is included in at least one aspect. Thus, appearances of the phrases "in one aspect," "in an aspect," "in an exemplification," and "in one exemplification" in various places throughout the specification are not necessarily all referring to the same aspect. Furthermore, the particular features, structures or characteristics may be combined in any suitable manner in one or more aspects.

[0115] As used herein, the singular form of "a", "an", and "the" include the plural references unless the context clearly dictates otherwise.

[0116] As used herein, the term "comprising" is not intended to be limiting, but may be a transitional term synonymous with "including," "containing," or "characterized by." The term "comprising" may thereby be inclusive or open-ended and does not exclude additional, unrecited elements or method steps when used in a claim. For instance, in describing a method, "comprising" indicates that the claim is open-ended and allows for additional steps. InAttorney Docket No. 3726023.00006 describing a device, "comprising" may mean that a named element(s) may be essential for an embodiment or aspect, but other elements may be added and still form a construct within the scope of a claim. In contrast, the transitional phrase "consisting of excludes any element, step, or ingredient not specified in a claim. This is consistent with the use of the term throughout the specification.

[0117] Any patent application, patent, non-patent publication, or other disclosure material referred to in this specification and / or listed in any Application Data Sheet is incorporated by reference herein, to the extent that the incorporated materials is not inconsistent herewith. As such, and to the extent necessary, the disclosure as explicitly set forth herein supersedes any conflicting material incorporated herein by reference. Any material, or portion thereof, that is said to be incorporated by reference herein, but which conflicts with existing definitions, statements, or other disclosure material set forth herein will only be incorporated to the extent that no conflict arises between that incorporated material and the existing disclosure material. None is admitted to be prior art.

[0118] In summary, numerous benefits have been described which result from employing the concepts described herein. The foregoing description of the one or more forms has been presented for purposes of illustration and description. It is not intended to be exhaustive or limiting to the precise form disclosed. Modifications or variations are possible in light of the above teachings. The one or more forms were chosen and described in order to illustrate principles and practical application to thereby enable one of ordinary skill in the art to utilize the various forms and with various modifications as are suited to the particular use contemplated. It is intended that the claims submitted herewith define the overall scope.

Claims

Attorney Docket No. 3726023.00006Claims:

1. A system deployed within a communication network, the system comprising:a computing device including:a non-transitory machine readable storage medium configured to store an application program, anda processor coupled to the non-transitory machine readable storage medium and configured to control a plurality of modules to execute instructions of the application program to obtain and transmit data associated with a patient working with a continuous glucose monitor (CGM);a computing server system configured to obtain the data associated with the patient working with the CGM and social determinants of health (SDoH) of the patient; anda machine learning (ML) based model, deployed, trained, and configured by the computing server system, to:receive the data associated with the patient working with the CGM and the SDoH of the patient,analyze patient behavior and attrition with the CGM over a selected period of time based at least upon the data associated with the patient working with the CGM and the SDoH of the patient,determine, in connection with an observation date, a plurality of predictive models for upcoming 1 to 3 months to identify a non-adherence event of the patient, determine a risk segment for the patient as a function of the patient behavior and attrition with the CGM over the selected period of time and results of the plurality of predictive models for upcoming 1 to 3 months in connection with the observation date, perform a cohort determination and a persona selection of the patient, and develop tailored intervention plans to engage patients for continuing adherence based on the cohort determination and the persona selection of the patient.

2. The system of claim 1, wherein the risk segment for the patient is one of: high risk attrition, medium risk attrition, high risk lapsing, medium risk lapsing, low risk, and adherent.

3. The system of claim 1, wherein the computing server system is further configured to process the data associated with the patient working with the CGM and the SDoH of the patient by removing variables relating to timestamp information, dates, identifier information, YAttorney Docket No. 3726023.00006 variables, and missing values, identifying a first portion of the data associated with the patient working with the CGM and the SDoH of the patient as categorical features, and identifying a second portion of the data associated with the patient working with the CGM and the SDoH of the patient as numerical features.

4. The system of claim 3, wherein, prior to running the ML based model, the computing server system is further configured to process the categorical features by removing data that are constant in nature, have high imbalance, contain missing data, have high cardinality, have high categorical association, and relate to data leakage and model tuning.

5. The system of claim 3, wherein, prior to running the ML based model, the computing server system is further configured to process the numerical features by removing data that contain missing values, have high correlation, have high multicollinearity with other variables, and relate to data leakage and model tuning, and adding selected features from the categorical features as numeric.

6. The system of claim 1, wherein the ML based model is further configured to perform the cohort determination of the patient based on the risk segment, an insurance type carried by the patient, and a binary indicator of the patient with respect to the CGM.

7. The system of claim 1, wherein the ML based model is further configured to perform the persona selection of the patient in accordance with a plurality of predefined persona groups created based at least on information relating to the patient’s demographic, psychographic and behavioral patterns, channel or engagement preferences, social determinants of health data, and diagnoses and chronic conditions.

8. The system of claim 1, wherein the ML based model is further configured to perform a Pearson chi-square testing to evaluate the tailored intervention plans in individual cohorts of Medicare, Medicaid, and commercial insurance plan enrollees who had lapsed or been lost to attrition.

9. The system of claim 1, wherein the ML based model is further configured to generate signals representing the tailored intervention plans and transmit the signals to the computing device for display.Attorney Docket No. 3726023.0000610. The system of claim 9, wherein the ML based model is further configured to determine a pl urali ty of channels for transmitting the signals to the computing device for display, wherein the plurality of channels include a combination of emails, text messages via short message / messaging service (SMS), and outbound interactive voice response.

11. A computing server system deployed within a communication network, the computing server system comprising:a non-transitory machine readable storage medium storing machine executable instructions; anda processor coupled to the non-transitory machine readable storage medium and configured to control a plurality of modules to execute the machine executable instructions to:obtain data relating to a patient, wherein the data include demographic information, psychographic and behavior patterns, previous recording channel of a continuous glucose monitor (CGM) and engagement preferences, social determinants of health (SDoH), and diabetes diagnoses and specific comorbidities; anddeploy and train a machine learning (ML) based model to:receive the data associated with the patient working with the CGM and the SDoH of the patient,analyze patient behavior and attrition with the CGM over a selected period of time based at least upon the data,determine, in connection with an observation date, a plurality of predictive models for upcoming 1 to 3 months to identify a non-adherence event of the patient,determine a risk segment for the patient as a function of the patient behavior and attrition with the CGM over the selected period of time and results of the plurality of predictive models for upcoming 1 to 3 months in connection with the observation date,perform a cohort determination and a persona selection of the patient, anddevelop tailored intervention plans to engage patients for continuing adherence based on the cohort determination and the persona selection of the patient.Attorney Docket No. 3726023.0000612. The computing server system of claim 11, wherein the risk segment for the patient is one of: high risk attrition, medium risk attrition, high risk lapsing, medium risk lapsing, low risk, and adherent.

13. The computing server system of claim 11, wherein the processor is further configured to control the plurality of modules to execute the machine executable instructions to process the data by removing variables relating to timestamp information, dates, identifier information, Y variables, and missing values, identifying a first portion of the data as categorical features, and identifying a second portion of the data as numerical features.

14. The computing server system of claim 13, wherein, prior to running the ML based model, the processor is further configured to control the plurality of modules to execute the machine executable instructions to process the categorical features by removing data that are constant in nature, have high imbalance, contain missing data, have high cardinality’, have high categorical association, and relate to data leakage and model tuning.

15. The computing server system of claim 13, wherein, prior to running the ML based model, the processor is further configured to control the plurality of modules to execute the machine executable instructions to process the numerical features by removing data that contain missing values, have high correlation, have high multicollinearity with other variables, and relate to data leakage and model tuning, and adding selected features from the categorical features as numeric.

16. The computing server system of claim 11, wherein the ML based model is configured to perform the cohort determination of the patient based on the risk segment, an insurance ty pe carried by the patient, and a binary indicator of the patient with respect to the CGM.

17. The computing server system of claim 11, wherein the ML based model is configured to perform the persona selection of the patient in accordance with a plurality of predefined persona groups created based at least on information relating to the patient’s demographic, psychographic and behavioral patterns, channel or engagement preference, social determinants of health data, and diagnoses and chronic conditions.Attorney Docket No. 3726023.0000618. The computing server system of claim 11, wherein the ML based model is configured to perform a Pearson chi-square testing to evaluate the tailored intervention plans in individual cohorts of Medicare, Medicaid, and commercial insurance plan enrollees who had lapsed or been lost to attrition.

19. The computing sen' er system of claim 11, wherein the ML based model is configured to generate signals representing the tailored intervention plans and transmit the signals to a computing device deployed within the communication network for display.

20. The computing server system of claim 19, wherein the ML based model is configured to determine a plurality of channels for transmitting the signals to the computing device for display, wherein the plurality of channels include a combination of emails, text messages via short message / messaging service (SMS), and outbound interactive voice response.

Images