US-12620485-B2 - Encoded graphical modeling system and method for matching patients or members needing a particular medical procedure or other health intervention with healthcare facilities or providers

US12620485B2US 12620485 B2US12620485 B2US 12620485B2US-12620485-B2

Abstract

A system and method for using a graph-based data structure to capture complex relations between different healthcare entities, analyze and mine healthcare data. The system sets up a framework for analyzing and mining historical healthcare data to help clinical patients and practitioners to guide care and make early decisions for interventions. More particularly, the system and method use graph embedding and machine learning modeling to process healthcare data in order to match member/patients with healthcare facilities or providers for performing a particular medical procedure or other health intervention needed by the patient.

Inventors

Yongjia Song
Peyman Yousefian
Rajiv Kumar Gumpina
Sravya Etlapur
Nataley Savanah Kennedy
Brandi Sambola
Rohan Vohra

Assignees

HUMANA INC.

Dates

Publication Date: 20260505
Application Date: 20221101

Claims (9)

1 . A system for predicting and recommending a particular healthcare facility or provider for those members or patients needing a particular medical procedure or other health intervention, the system comprising: a database for storing historical claims data; a computer processor; a graphical user interface comprised of a first region for entering a zip code, and a second region for listing recommended healthcare facilities or providers for performing the particular medical procedure or other health intervention based on the entered zip code; a non-transitory computer-readable medium storing instructions that when executed by the computer processor cause the computing device to perform the steps of: a. extracting a heterogeneous graph from the historical claims data, the graph comprised of at least two types of nodes selected from the group comprising a member node, a healthcare facility node, and a provider node, the heterogeneous graph further comprising a plurality of edges each representing a positive class, and a plurality of edges each representing a negative class, wherein each edge representing a positive class connects one member node to one healthcare facility node or provider node that the one member has a previous connection or visit with and wherein each edge representing a negative class connects one member node to one healthcare facility node or provider node that the one member does not have a previous connection or visit with; b. embedding the heterogeneous graph to generate a member node vector for each member node and a healthcare facility node vector for each healthcare facility node or a provider node vector for each provider node and sampling a neighborhood for each member node and healthcare facility node or provider node and aggregating node feature information from the sampled neighborhood to update each member node vector and each healthcare facility node vector or provider node vector, concatenating a plurality of vectors, each concatenated vector comprised of a first member node vector and at least one healthcare facility node vector or at least one provider node vector, wherein each of the plurality of concatenated vectors represents either a member-healthcare facility pair or member-provider pair; c. applying a deep learning model to the plurality of concatenated vectors to determine the probability that each member-healthcare facility pair or member-provider pair will be selected for the particular medical procedure or other health intervention, the deep learning model having been trained using labeled member-healthcare-facility pairs or member-provider pairs derived from historical claims data, assigning a positive label where a member previously visited the facility or provider and a negative label where the member did not, and updating model weights by minimizing a loss function; and populating the second region with the recommended healthcare facilities or providers with the highest probabilities for performing the particular medical procedure or other health intervention based on an entered zip code for the patient or member.
2 . The system according to claim 1 , wherein the heterogeneous graph is further comprised of: member node features including age, gender and medical conditions; and healthcare facility node features including facility performance and quality.
3 . The system according to claim 1 , further comprising: a first model used to score the likelihood that each member would receive the particular medical procedure or other health intervention at the particular type of healthcare facility or with one of the providers before determining the probability that each member-healthcare facility pair or member-provider pair will be selected for the particular medical procedure or other health intervention.
4 . The system according to claim 1 , wherein the non-transitory computer-readable medium stores instructions that when executed by the computer processor cause the computing device to perform, prior to extracting the graph, a step of generating the plurality of edges where each edge represents a negative class, this step further comprising the steps of: assigning all available healthcare facilities or providers to each member; determining a distance between each member location and each healthcare facility location or provider location; grouping each distance between each member location and each healthcare facility location or provider location into a predetermined number of categories based on the determined distance; matching each edge representing a positive class with one randomly selected edge representing a negative class from each of the categories; and preparing the graph.
5 . The system according to claim 1 , wherein the particular medical procedure needed is Esophagogastroduodenoscopy (EGD) and the particular healthcare facility is an Ambulatory Surgery Center (ASC).
6 . The system according to claim 1 , wherein a weight matrix is applied to the plurality of concatenated vectors to produce final probability scores.
7 . The system according to claim 1 , wherein the heterogeneous graph is comprised of a training set comprised of a predetermined number of randomly selected member nodes and a testing set comprised of the rest of the member nodes that were not randomly selected for the training set.
8 . The system according to claim 1 , wherein the graphical user interface is further comprised of a third region for entering in member identification information and a fourth region for entering in a minimum threshold distance that the member-healthcare facility pair or member-provider must fall below.
9 . The system according to claim 1 , wherein the non-transitory computer-readable medium stores instructions that when executed by the computer processor cause the computing device to generate a neighborhood embedding and concatenate the neighborhood embedding with a respective node vector to update each member node vector and each healthcare facility node vector or provider node vector.

Description

BACKGROUND OF THE INVENTIVE FIELD The present invention is directed to a system and method for using a graph-based framework and structure to capture complex relations between different healthcare entities. The analysis and data mining of rich amounts of interactions between healthcare entities can help clinical practitioners to guide care and make early decisions for interventions. More particularly, the present invention uses graph embedding and machine learning to process healthcare data in order to match member/patients with healthcare facilities or healthcare providers for performing a particular medical procedure or other health intervention needed by the member/patient. The standardization and sharing of massive amounts of healthcare data enables data-driven analysis using machine learning to solve healthcare problems. In recent years, massive healthcare artificial intelligence (AI) applications have been proposed such as the prediction of medical conditions and disease. In order to maximize the benefits of AI models and capture the relationships between multiple health entities, a graph structure should be established so that graph embedding can then be applied. Healthcare data usually includes administrative claims, demographics information, diagnosis, conditions, treatments, prescriptions, provider information, hospitalization, insurance, etc. Various entities are involved under healthcare settings: patients, physicians, hospital, other health facility, etc. Building the connection between these entities will play a significant role in lowering down data storage and organization efforts and shedding light in providing better answers for solving healthcare related questions or problems. Graph analytics, also called network analysis, is the analysis of relationships among multiple entities. In recent years, graph analytics has been applied in various areas, such as resource management, fraud detection, social network analysis, etc. A graph is composed of a set of nodes and a set of edges. FIG. 1 illustrates an example graph comprised of nodes and edges. Nodes can represent different entities, and edges can represent the relationship between a pair of nodes. Graph embedding is a technique to transform a graph to a vector or set of vectors. It can capture the information of graph topology, node attributes and neighborhood attributes. One key advantage of applying graph embedding is to convert a complex graph data model into a lower dimensional space which can maximally preserve graph structure and information. The embedding process converts particular input data to be analyzed into a computer-readable vector format. Once embedded, learning for tasks such as disease prediction can be carried out. The particulars of the embedding process has a significant impact on the performance of the model for future analysis and tasks, and it is important for a quality model to be developed that reflects the accuracy of the data and relationships. In one practical application of the invention, the framework and processes of the present invention may be used to predict the possibility of clinicians choosing a particular healthcare facility or provider for patients needing a particular medical procedure (e.g., Esophagogastroduodenoscopy (EGD)) and to provide a recommendation for a particular healthcare facility (e.g., hospital, Ambulatory Surgery Center (ASC), clinic, urgent care centers, etc.) or provider. SUMMARY OF THE GENERAL INVENTIVE CONCEPT In one embodiment of the invention, the invention is comprised of: a system for predicting and recommending a particular healthcare facility or provider for those members or patients needing a particular medical procedure or other health intervention, the system comprising: a database for storing historical claims data; a plurality of concatenated vectors each comprised of a first member vector and either a first healthcare facility vector or provider vector concatenated together, wherein each of the plurality of concatenated vectors represents either a member-healthcare facility pair or member-provider pair; a computer processor; a non-transitory computer-readable medium storing instructions that when executed by the computer processor cause the computing device to perform the steps of: a. applying a deep learning model to the plurality of concatenated vectors; andb. determining the probability that each member-healthcare facility pair or member-provider pair will be selected for the particular medical procedure or other health intervention. In one embodiment, the graph is further comprised of: a heterogeneous graph extracted from the historical claims data, the graph comprised of at least two types of nodes selected from the group comprising a member node, a healthcare facility node, and a provider node, the heterogeneous graph further comprising a plurality of edges each representing a positive class, and a plurality of edges each representing a negative class, wherein e