# US 2022/0093268 A1 · Cross-departmental chronic kidney disease early diagnosis and decision support system based on knowledge graph

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2022/0093268 A1
- Document ID: US20220093268A1
- Status: Published application
- Kind: A1
- Application number: 17/541,301
- Filed: 2021-12-03
- Published: 2022-03-24
- Earliest claimed: 2020-01-14
- Assignees: Zhejiang Lab
- Applicants: Zhejiang Lab
- Inventors: Jingsong LI; Yu TIAN; Yong SHANG; Ran XIN
- Web page: https://patentgrep.com/patent/US20220093268A1

## CPC

- G16H 50/50: for simulation or modelling of medical disorders
- G16H 10/60: for patient-specific data, e.g. for electronic patient records
- G16H 50/20: for computer-aided diagnosis, e.g. based on medical expert systems
- G16H 50/70: for mining of medical data, e.g. analysing previous cases of other patients
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA

## Abstract

Provided is a cross-departmental decision support system for early diagnosis of a chronic kidney disease based on knowledge graph, which comprises a patient information model building module, a patient information model library storage module, a knowledge graph association module, a knowledge graph inference module and a decision support feedback module. According to the present application, by constructing a patient information model and utilizing an OMOP CDM standard terminology system, patient electronic medical record data is constructed into a patient information model with unified concept coding and unified semantic structure; making full use the advantages of semantic technology in data interactivity and scalability, so that the system has better adaptability and scalability to heterogeneous data in different hospitals.

## Claims

### Claim 1

A cross-departmental decision support system for early diagnosis of a chronic kidney disease based on knowledge graph, comprising a patient information model building module, a patient information model library storage module, a knowledge graph association module, a knowledge graph inference module and a decision support feedback module; wherein
the patient information model building module is configured to build a patient-centered patient electronic medical record data knowledge sub-graph based on electronic medical record data of a patient, according to a semantic structure of a chronic kidney disease knowledge graph and an OMOP CDM data structure to form a patient information model;
the patient information model library storage module provides support for storage and calling of the patient information mode, stores the patient information model in a RDF triple form by establishing a knowledge graph database, and provides a query and modification interface of the patient information model by a SPARQL endpoint; construct a SPARQL query statement template according to content and structure of patient data needed for early diagnosis of the chronic kidney disease, and provides a general method for retrieving the patient information model, so as to support decision of the knowledge graph to support inference;
the knowledge graph association module enriches semantic information of the patient information model, and associates scattered clinical record RDF triples in the patient information model according to a real world clinical diagnosis and treatment process sequence; associates examples of medical examinations, clinical complaints and prescription records with clinical discoveries, clinical diagnoses and disease risks based on a medical concept relationship in the chronic kidney disease knowledge graph;
the knowledge graph inference module reasons diagnosis deficiency, disease risk diagnosis deficiency, disease follow-up requirement deficiency and prescription adjustment for the chronic kidney disease of a patient based on the RDF triple relationship of clinical records in the patient information model and clinical recommendation inference rules in the chronic kidney disease knowledge graph, and provides corresponding clinical diagnosis recommendations, risk recommendations, follow-up requirement recommendations and prescription adjustment recommendations based on inference results; and
the decision support feedback module feeds back the clinical diagnosis recommendations, risk recommendations, follow-up requirement recommendations and prescription adjustment recommendations provided by the knowledge graph inference module, and a clinical recommendation report containing inference process of above recommendations to a physician.

### Claim 2 (depends on 1)

The cross-departmental decision support system for early diagnosis of a chronic kidney disease based on knowledge graph according to claim 1, wherein in the patient information model building module, model building comprises two processes which are patient data analysis and patient data RDF semantic transformation:
(1) patient data analysis process: according to the electronic medical record data of the patient, the patient data is divided into multiple hierarchical structures according to the chronic kidney disease knowledge graph and the standard terminology coding and structure of OMOP CDM; for each patient P, there are multiple medical records Vi(i=1 . . . n), where {Vi|i=1 . . . n}⊂P; wherein Vi∩Vj=ø, (Vi, Vj)⊂P; each visit record Vi spans a time period T, and for information such as any piece of diagnosis record D, examination record M, prescription record Pr and operation record Pc in the patient data, {Dh,Mj,Prk,Pcm|h,j,k,m=1 . . . n}⊂Vi when a record time thereof is within the time period T; the concept coding, relationship type, value type and value information are extracted from subitems of D, M, Pr and Pc; and
(2) a RDF semantic transformation process of the patient data: based on the analysis results of the patient data and the semantic structure of the chronic kidney disease knowledge graph, the patient electronic medical record data is constructed into a RDF triple relationship conforming to OWL language specification, and a data field is converted to standard OMOP CDM terminology coding through semantic mapping; for each patient P, medical record V, examination record M and prescription record Pr in the patient data, an ontology instance is constructed with a data ID thereof as a URI to form a corresponding instance diagram; for each patient instance Pind, medical record instance Vind, examination record instance Mind and prescription record instance Prind, a corresponding ontology class relationship is constructed to form a corresponding class diagram; based on a hierarchical relationship intercepted in patient data analysis, association between the patient instance Pind and the medical record instance Vind, association between the medical record instance Vind and the examination record instance Mind, and association between the medical record instance Vind and the prescription record instance Prind are established through ontology object properties to form a patient data relationship diagram and a medical record relationship diagram; for each diagnostic record D and operation record Pc, a relationship between the related medical record instance Vind and a disease ontology class and an operation record ontology class is established to form a corresponding class diagram; the patient data is constructed into RDF triple data according to the above rules, and the patient information model is formed.

### Claim 3 (depends on 1)

The cross-departmental decision support system for early diagnosis of a chronic kidney disease based on knowledge graph according to claim 1, wherein the patient information model library storage module is divided into three parts: a Jena TDB storage terminal, a Jena Fuseki SPARQL endpoint and a control terminal;
the Jena TDB storage terminal is configured to build a knowledge graph database to store the patient information model in a form of RDF triples, wherein all patient information models converted into the RDF triples are recorded as N-Triple data files conforming to OWL language rules in a form of <s,p,o>, a RDF triple data set is stored and transformed by a Jena TDB Loader, and indices of a head element S, a relation element P and a tail element O of the RDF triple is established to improve the query speed;
the Jena Fuseki SPARQL endpoint is configured to build a RDF triple query and modification interface; the Jena Fuseki SPARQL endpoint invokes Jena TDB storage content, and invokes and modifies the patient information model through a W3C standard SPARQL query statement; the Jena Fuseki SPARQL endpoint is built in a Tomcat server, which monitors a SPARQL query request through a HTTP request and feeds back corresponding results;
the control terminal provides a target query and modification function for the patient information model; according to patient information model structure and knowledge graph inference requirements, a SPARQL query statement template is preset, and a request is sent to the Jena Fuseki SPARQL endpoint and the output result is formatted; for each patient Pind, according to the clinical records in the patient information model, a result of a patient data relationship diagram is queried; based on the medical record example Vind that has been fed back, a result of the medical record relationship diagram is queried; finally, the data property information is queried according to the examination record instance Mind and the operation record instance Prind; the control terminal sets SPARQL query statements based on above principles, and establishes a result graph for feedback.

### Claim 4 (depends on 1)

The cross-departmental decision support system for early diagnosis of a chronic kidney disease based on knowledge graph according to claim 1, wherein the knowledge graph association module is implemented as follows:
the chronic kidney disease knowledge graph is defined as G=(V,A), and the patient information model is defined as G′=(V′,A′), where G and G′ are two directed graphs, V and V′ are nodes in the graph, and A and A′ are directed edges in the graph; the knowledge graph association module completes V′ and A′ of graph G′ based on the relationship in graph G; for any v∈V and any v′∈V′, node similarities sim(v,v′) and sim′(v′,v) are calculated through cos similarity; at the same time, similarity matching is carried out according to a standard concept coding hierarchy relationship between nodes v and v′; a similar node pair is recorded as b=min|sim(vi,vj′)−sim′(vj′,vi)|, a set thereof is b∈B, and the node pairs vi and vj contained in B are candidate associated node pairs; for the candidate association node pairs, the semantic association similarity of the nodes is calculated, and the triple relationship is supplemented to the patient information model according to the similar association, so as to improve the information; a triple (vi,a,vj)∈V×A×V is established for node v and directed edge A, where (vi,a,vj) conforms to graph relationship of G=(V,A); the calculation of semantic association is recorded as rel(v,a)={x|v,x∈V∧(v,a,x)∈A}; for each patient information model node vi′, sim(v,vi) of the corresponding knowledge graph node v is calculated, where (v,v)∈V and vi≠v; all node pairs v and vi whose sim(v,vi) is greater than a threshold k are taken, and if v and vi conform to rel(v,a) relationship, a <vi′,a,vi> triple association to the node v1′ of the patient information model is added, thereby realizing semantic information completion of the patient information model based on the chronic kidney disease knowledge graph.

### Claim 5 (depends on 1)

The cross-departmental decision support system for early diagnosis of a chronic kidney disease based on knowledge graph according to claim 1, wherein the knowledge graph inference module is implemented as follows:
firstly, the patient information model is invoked to analyze the RDF triple relationship therein; at the same time, an inference ontology copy On is established from the chronic kidney disease knowledge graph, and according to each instance S and an ontology class C to which the instance S belongs in the patient information model, the instance is constructed into an ontology class C′ corresponding to the inference ontology copy On; each instance S and a relation graph R thereof in the patient information model are extracted to obtain a <si,r,sj>triple, where (si,sj)∈S,r∈R, which is reconstructed in the inference ontology copy on; each instance S and a data property relationship Rs and an attribute Value Va thereof in the patient information model are extracted to obtain a <s,rs,va> triple, where s ∈ S,rs ∈ Rs, va ∈ Va, which is reconstructed in the inference ontology copy On, in the above steps, the inference ontology copy On of chronic kidney disease knowledge graph containing patient information is constructed to provide basic elements for semantic inference;
then, based on OWL2 DL rules in the chronic kidney disease knowledge graph, by using the Fact++ inference engine, according to the patient information in the inference ontology copy On, the patient instance SP. the medical record instance SV, the examination record instance SM and the prescription record instance SPr are subjected to ontology class affiliation completion and missing object property establishment; according to OWL2 DL rules in the chronic kidney disease knowledge graph, the data in the patient information model, such as examination record examples of abnormal glomerular filtration rate, prescription record examples containing drugs that may damage renal function, and medical record examples containing chief complaints of chronic kidney disease symptoms, are associated with the chronic kidney disease risk ontology class; object property association is established for the medical record instance Sv′ including abnormal examination record instance SM′ and prescription record instance SPr′, and abnormal records in the medical records are marked;
according to Jean Rules established in the clinical guidelines for chronic kidney diseases, by using Jena inference engine, the association between a patient instance SP and a medical record instance Sv and a clinical recommendation instance SP such as disease diagnosis, risk factors, follow-up plans and drug adjustment plans in the chronic kidney disease knowledge graph is established through semantic inference based on the instance data and the instance relationship in the patient information model, thus forming clinical recommendation contents based on the patient's condition; the Jean Rules established in the clinical guidelines for chronic kidney diseases are divided into diagnosis rules, risk classification rules, follow-up screening rules and prescription adjustment rules for chronic kidney diseases; above rules are compiled based on a Jena Rules standard language, and are established according to the paradigm of [rule name: (triple 1) (triple 2) . . . operator 1 (operand) operator 2 (operand) . . . →(triple N) (triple N+1) . . . ]; according to the specifications of diagnosis, risk grading, follow-up screening and prescription adjustment for chronic kidney diseases the in clinical guidelines, RDF triples containing medical examination results, chief complaint symptoms, disease history, prescription and other information of the patient are established on the left side of the arrow as conditions, a numerical value of that examination result is judged by operators and operands, and RDF triples of disease diagnosis, risk classification, follow-up screening and prescription adjustment are established on the right side of the arrow when the conditions on the left side of the arrow are met; the Jena inference engine implements inference according to the RDF triple information corresponding to the patient instance SP in the patient information model based on the requirements on the left side of the arrow, and adds a decision support recommendation RDF triple on the right side of the arrow to the patient instance Sp that meets the rules;
according to the inference rules triggered by a Fact++ inference engine and a Jena inference engine, the inference instance SR is established, and the inference instance SR is configured to record the clinical knowledge in the clinical guidelines involved in inference; attributes of a relation between the inference instance SR and the medical record instance SV, the examination record instance SM and the prescription record SPr involved in the inference rules are established, so as to establish inference result association for the clinical recommendations and provide a recommendation reason inference process record.

### Claim 6 (depends on 1)

The cross-departmental decision support system for early diagnosis of a chronic kidney disease based on knowledge graph according to claim 1, wherein the decision support feedback module is specifically implemented as follows:
the class, object property and data property content of the patient instance SP in a regular inference ontology On are queried through a preset SPARQL query statement, the obtained RDF graph results are analyzed, and the relevant information (concept coding, concept name, numerical value, text information, etc.) of tail nodes in the RDF triple are extracted, so as to obtain a newly established example Sre of diagnosis, risk, follow-up recommendations and prescription adjustment recommendations for the chronic kidney disease; the diagnosis, the risk, the follow-up recommendations and the prescription adjustment recommendations for the chronic kidney diseases that are newly established by the inference module are obtained from Sre, and meanwhile a clinical recommendation report containing inference process is formed by combining an inference path relationship recorded in the inference instance SR corresponding to the patient instance SP and the recommendation instance SRE for feedback to physicians.
