Systems and methods for generating personalized care paths for patients

1 . A computer-implemented method comprising:

receiving, by one or more processors, a plurality of training datasets associated with a plurality of patients, the plurality of training datasets including (i) a plurality of relevant data associated with the plurality of patients including demographic data and medical data associated with the plurality of patients and (ii) a plurality of known care paths for the plurality of patients, wherein each known care path of the plurality of known care paths includes one or more known procedures, from a plurality of procedures, undergone by a respective patient;

generating, by the one or more processors and based on the plurality of training datasets, a patient-bucket-procedure (PBP) graph, the PBP graph including:

(i) a plurality of nodes comprising (a) a plurality of patient nodes representing the plurality of relevant data associated with the plurality of patients, (b) a plurality of procedure code nodes representing a plurality of procedure codes for the plurality of procedures, and (c) a plurality of bucket nodes for the plurality of patients that represent the plurality of known care paths for the plurality of patients, and

(ii) a plurality of links comprising (a) first links between the plurality of bucket nodes and the plurality of patient nodes corresponding to the plurality of patients, and (b) second links between the plurality of bucket nodes and the plurality of procedure code nodes representing procedure codes for the one or more known procedures;

defining, by the one or more processors and based on a structure of the PBP graph, an architecture of a graph convolutional neural network-based model, the architecture including (i) an input layer configured to receive input data, (ii) a plurality of embedding layers configured for use with a plurality of heterogenous aggregators to learn an embedding of each node of the plurality of patient nodes, the plurality of bucket nodes, and the plurality of procedure code nodes by modeling interactive signals, and (iii) an output layer configured to generate output data based on the learned embeddings;

training, by the one or more processors using the PBP graph and based on the plurality of training datasets, the graph convolutional neural network-based model;

receiving, by the one or more processors and via the input layer of the trained graph convolutional neural network-based model, new relevant data associated with a new patient including demographic data and medical data associated with the new patient as the input data;

identifying, by the one or more processors and via the plurality of embedding layers of the trained graph convolutional neural network-based model, (i) a patient from the plurality of patients similar to the new patient and (ii) one or more procedures represented by one of the plurality of bucket nodes corresponding to the patient;

generating, by the one or more processors and via the output layer of the trained graph convolutional neural network-based model, a personalized care path for the new patient as the output data, the personalized care path including the one or more procedures identified; and

providing, by the one or more processors, data associated with the personalized care path for the new patient to a device associated with a user.

2 . The computer-implemented method of claim 1 , wherein the plurality of procedure code nodes include a plurality of current procedural terminology (CPT) codes corresponding to the plurality of procedures, and wherein the plurality of bucket nodes are determined by:

determining, by the one or more processors, the one or more known procedures included in the plurality of known care paths, the one or more known procedures occurring during a pre-determined time period;

determining, by the one or more processors, one or more of the plurality of CPT codes corresponding to the one or more known procedures; and

ordering, by the one or more processors, the one or more of the plurality of CPT codes chronologically within the plurality of bucket nodes based on an occurrence of the one or more known procedures during the pre-determined time period.

3 . The computer-implemented method of claim 2 , wherein each of the plurality of CPT codes is included in one or more of the plurality of bucket nodes.

4 . The computer-implemented method of claim 2 , wherein each of the plurality of bucket nodes includes a set of CPT codes, from the plurality of CPT codes, that are semantically related.

5 . The computer-implemented method of claim 1 , wherein the plurality of heterogenous aggregators of the graph convolutional neural network-based model include:

a patient aggregator configured to learn relatedness between the plurality of patient nodes;

a bucket aggregator configured to learn relatedness between the plurality of bucket nodes; and

a CPT aggregator configured to learn relatedness between the plurality of procedure code nodes.

6 . The computer-implemented method of claim 1 , wherein at least one known procedure from the one or more known procedures included in the plurality of known care paths is part of a treatment plan for medical conditions of the plurality of patients, and wherein the one or more procedures identified and included in the personalized care path are to be performed with respect to the new patient to treat a medical condition of the new patient.

7 . The computer-implemented method of claim 1 , wherein the demographic data includes at least one of: age information, gender information, location information, income level, education level, household data, ethnic origin, employment data, marital status, children data, or languages spoken.

8 . The computer-implemented method of claim 1 , wherein the medical data includes at least one of: health condition, claim information, medication information, hospitalization information, lab information, or procedures information.

9 . The computer-implemented method of claim 1 , further comprising:

causing to be displayed, by the one or more processors, at least a portion of the data associated with the personalized care path for the new patient via a graphical user interface (GUI) on the device.

10 . The computer-implemented method of claim 9 , wherein the at least the portion of the data associated with the personalized care path for the new patient includes at least one of:

the one or more procedures identified and included in the personalized care path;

a transaction variable, including a cost estimate, associated with the personalized care path; or

a medical variable, including a duration or an intensity, associated with the personalized care path.

11 . A system comprising:

one or more processors; and

one or more non-transitory computer readable media storing processor-executable instructions which, when executed by the one or more processors, cause the one or more processors to perform operations comprising:

receiving a plurality of training datasets associated with a plurality of patients, the plurality of training datasets including (i) a plurality of relevant data associated with the plurality of patients including demographic data and medical data associated with the plurality of patients and (ii) a plurality of known care paths for the plurality of patients, wherein each known care path of the plurality of known care paths includes one or more known procedures, from a plurality of procedures, undergone by a respective patient;

generating, based on the plurality of training datasets, a patient-bucket-procedure (PBP) graph, the PBP graph including:

(i) a plurality of nodes comprising (a) a plurality of patient nodes representing the plurality of relevant data associated with the plurality of patients, (b) a plurality of procedure code nodes representing a plurality of procedure codes for the plurality of procedures, and (c) a plurality of bucket nodes for the plurality of patients that represent the plurality of known care paths for the plurality of patients, and

(ii) a plurality of links comprising (a) first links between the plurality of bucket nodes and the plurality of patient nodes corresponding to the plurality of patients, and (b) second links between the plurality of bucket nodes and the plurality of procedure code nodes representing procedure codes for the one or more known procedures;

defining, based on a structure of the PBP graph, an architecture of a graph convolutional neural network-based model, wherein the architecture includes (i) an input layer configured to receive input data, (ii) a plurality of embedding layers configured for use with a plurality of heterogenous aggregators to learn an embedding of each node of the plurality of patient nodes, the plurality of bucket nodes, and the plurality of procedure code nodes by modeling interactive signals, and (iii) an output layer configured to generate output data based on the learned embeddings;

training, using the PBP graph and based on the plurality of training datasets, the graph convolutional neural network-based model;

receiving, via the input layer of the trained graph convolutional neural network-based model, new relevant data associated with a new patient including demographic data and medical data associated with the new patient as the input data;

identifying, via the plurality of embedding layers of the trained graph convolutional neural network-based model, (i) a patient from the plurality of patients similar to the new patient and (ii) one or more procedures represented by one of the plurality of bucket nodes corresponding to the patient;

generating, via the output layer of the trained graph convolutional neural network-based model, a personalized care path for the new patient as the output data, the personalized care path including the one or more procedures identified; and

providing data associated with the personalized care path for the new patient to a device associated with a user.

12 . The system of claim 11 , wherein the plurality of procedure code nodes include a plurality of current procedural terminology (CPT) codes corresponding the plurality of procedures, and wherein the plurality of bucket nodes are determined by:

determining the one or more known procedures included in the plurality of known care paths, the one or more known procedures occurring during a pre-determined time period;

determining one or more of the plurality of CPT codes corresponding to the one or more known procedures; and

ordering the one or more of the plurality of CPT codes chronologically within the plurality of bucket nodes based on an occurrence of the one or more known procedures during the pre-determined time period.

13 . The system of claim 12 , wherein each of the plurality of CPT codes is included in one or more of the plurality of bucket nodes.

14 . One or more non-transitory computer readable media storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:

receiving a plurality of training datasets associated with a plurality of patients, the plurality of training datasets including (i) a plurality of relevant data associated with the plurality of patients including demographic data and medical data associated with the plurality of patients and (ii) a plurality of known care paths for the plurality of patients, wherein each known care path of the plurality of known care paths includes one or more known procedures, from a plurality of procedures, undergone by a respective patient;

generating, based on the plurality of training datasets, a patient-bucket-procedure (PBP) graph, the PBP graph including:

(i) a plurality of nodes comprising (a) a plurality of patient nodes representing the plurality of relevant data associated with the plurality of patients, (b) a plurality of procedure code nodes representing a plurality of procedure codes for the plurality of procedures, and (c) a plurality of bucket nodes for the plurality of patients that represent the plurality of known care paths for the plurality of patients, and

(ii) a plurality of links comprising (a) first links between the plurality of bucket nodes and the plurality of patient nodes corresponding to the plurality of patients, and (b) second links between the plurality of bucket nodes and the plurality of procedure code nodes representing procedure codes for the one or more known procedures;

defining, based on a structure of the PBP graph, an architecture of a graph convolutional neural network-based model, wherein the architecture includes (i) an input layer configured to receive input data, (ii) a plurality of embedding layers configured for use with a plurality of heterogenous aggregators to learn an embedding of each node of the plurality of patient nodes, the plurality of bucket nodes, and the plurality of procedure code nodes by modeling interactive signals, and (iii) an output layer configured to generate output data based on the learned embeddings;

training, using the PBP graph and based on the plurality of training datasets, the graph convolutional neural network-based model;

receiving, via the input layer of the trained graph convolutional neural network-based model, new relevant data associated with a new patient including demographic data and medical data associated with the new patient as the input data;

identifying, via the plurality of embedding layers of the trained graph convolutional neural network-based model, (i) a patient from the plurality of patients similar to the new patient and (ii) one or more procedures represented by one of the plurality of bucket nodes corresponding to the patient;

generating, via the output layer of the trained graph convolutional neural network-based model, a personalized care path for the new patient as the output data, the personalized care path including the one or more procedures identified; and

providing data associated with the personalized care path for the new patient to a device associated with a user.

15 . The computer-implemented method of claim 1 , wherein the plurality of relevant data associated with the plurality of patients that is represented by the plurality of patient nodes of the PBP graph includes a plurality of variables, and the learned embeddings used by the graph convolutional neural network-based model include a learned relationship between the plurality of variables and the one or more known procedures, from the plurality of procedures, undergone by the plurality of patients.

16 . The computer-implemented method of claim 1 , wherein the graph convolutional neural network-based model uses the learned embeddings as a graph encoder.

17 . The computer-implemented method of claim 1 , further comprising:

determining, by the one or more processors, a transaction variable associated with the personalized care path based on the one or more procedures included in the personalized care path and a portion of the new relevant data associated with the new patient; and

replacing, by the one or more processors, and based on the determined transaction variable, at least one of the one or more procedures in the personalized care path with an alternative procedure.

18 . The computer-implemented method of claim 1 , wherein modeling the interactive signals includes building, based on the PBP graph, a patient-bucket interaction matrix, a bucket-procedure code interaction matrix, and a patient-procedure code interaction matrix.

19 . The computer-implemented method of claim 1 , wherein the plurality of embedding layers include: a first embedding layer configured to model patient-bucket node interactive signals based on the first links, a second embedding layer configured to model bucket-procedure code node interactive signals based on the second links, and a third embedding layer configured to model patient-procedure code node interactive signals based on the first links and the second links.

20 . The computer-implemented method of claim 1 , wherein generating the PBP graph comprises:

for a bucket node of the plurality of bucket nodes that represents a known care path, from the plurality of known care paths, having multiple known procedures, (i) categorizing the known care path into multiple segments based on procedure characteristics and a time of occurrence of the multiple known procedures over a time period spanning the known care path, and (ii) using a natural language processing (NLP) model to build a semantic relationship between the multiple segments based on procedure code descriptions for the multiple known procedures and the time of occurrence,

wherein, based on the semantic relationship built, the graph convolutional neural network-based model is trained to predict whether certain procedures occur together within a pre-defined time window of each other.