Data source classification, tracing and recommendation method for clinical research data

Abstract

The invention discloses a data source classification, tracing and recommendation method for clinical research data, and belongs to the field of medical data processing. The invention discloses a data source classification, tracing and recommendation method for clinical research data. The method comprises the following steps: acquiring clinical research case report form problems; converting a table name and a corresponding data field in the electronic medical record report table into metadata text description; the metadata text description is classified into a plurality of large classes, and ETL data transmission programs and codes used for transmitting source data are generated; obtaining a desensitized source data sample; searching and ranking source data categories for each research problem; a traceability path is determined through an artificial rechecking module, and a data interaction specification form is generated; by means of the arrangement, sensitive patient data used in clinical research can be effectively protected, and meanwhile the data transmission and processing efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the field of medical data processing, and more specifically, relates to a method for data source classification, tracing and recommendation of clinical research data. Background Art

[0002] In the process of clinical research, data sources are diverse and complex, such as electronic medical record systems (EMR), laboratory test systems (LIS) and drug management systems. These data are usually stored in different systems and in different formats, making data management, analysis and traceability extremely complicated. Current technologies mainly rely on manual operations, which are inefficient and prone to errors, and cannot meet the needs of modern clinical research for efficient and accurate data management.

[0003] In addition, according to relevant regulations, researchers can only use patient clinical data within the necessary scope when conducting clinical research to reduce the risk of data leakage and exposure of sensitive information; patient clinical data are generally stored in the hospital's internal system, which makes it impossible for researchers to effectively determine what specific data the hospital can provide for research projects. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method for data source classification, tracing and recommendation for clinical research data, which can effectively protect sensitive patient data used in clinical research while improving the efficiency of data transmission and processing.

[0005] A method for data source classification, tracing and recommendation of clinical research data of the present invention comprises the following steps:

[0006] S1: Access to clinical study case report form questions;

[0007] S2: Recommended categories of relevant hospital source data for answering questions about pathology reports, including:

[0008] S201: converting the table name and corresponding data fields of the electronic medical record report form obtained in S1 into metadata text description;

[0009] S202: Use semantic embedding technology and HNSW vector library retrieval technology to classify metadata text descriptions; embed metadata into semantic vector representation using Embedding technology, and then use HNSW algorithm ideas to build a database; use semantic similarity retrieval technology to achieve rapid matching of data fields with metadata and their data sources;

[0010] S203: Using the combination of Embedding technology and HNSW algorithm, a data world model is constructed to process unstructured and structured metadata respectively;

[0011] S3: Generate ETL data transmission program and code for transmitting source data; specifically, it includes:

[0012] S301: applying the algorithm of S2 to pre-process the unstructured metadata and / or structured metadata while retaining the corresponding data source;

[0013] S302: Generate a data mapping document, which includes sample SQL codes required for each type of metadata and a description of the use of metadata in the case report form;

[0014] S4: Obtain anonymized sample source data;

[0015] S5: Search and sort source data categories for each research question;

[0016] S6: Manually determine the traceability path;

[0017] S7: Generate a data interaction specification form.

[0018] As a further improvement of the present invention, in step S203, the unstructured metadata is structured into an architecture with patient ID, consultation ID, recording time, source data name, text content, additional field name and additional field result; other content stored in the hospital data system should be linked and sent to the additional field result, and metadata corresponding to the additional content is added to the additional field name to describe the result.

[0019] As a further improvement of the present invention, in step S203, the structured metadata is structured into the following architecture: patient ID, consultation ID, recording time, source data name, specified data field name, specified field result, additional field name, additional field result; for structured data fields, common data fields are used as specified fields, and additional fields cover other data stored in the hospital data system.

[0020] As a further improvement of the present invention, in step S5, each research question in the case report form is traversed, and the N text segments in the sample library that are most relevant to the research question and the data sources corresponding to the text segments are matched using the technology of S202; then the frequency of occurrence of each data source is counted and sorted in descending order according to the frequency, and the top three data sources with the highest evaluation rate are retained as the recommended data sources for the research question;

[0021] The research question is the research field in the case report form.

[0022] As a further improvement of the present invention, N is a user-defined value.

[0023] As a further improvement of the present invention, the value of N is 10.

[0024] As a further improvement of the present invention, in step S6, a manual review module is provided, and the manual review module is used to review and select the data source related to each research question in the case report form.

[0025] As a further improvement of the present invention, in step S7, the result form in S302 is updated using the data source after manual review in S6 as the final data interaction standard form.

[0026] As a further improvement of the present invention, in step S4, the data mapping document in S302 is desensitized to obtain anonymized sample source data after desensitization.

[0027] As a further improvement of the present invention, each metadata has a corresponding data field structure.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention can recommend the data source category corresponding to the specific data requirements of clinical research, and generate a data transmission requirement table (such as laboratory results from the LIS system); this not only improves the review and efficiency of data transmission, but also ensures that the transmitted data is de-identified patient sample data, thereby further protecting patient privacy; in addition, the present invention can classify and sort the data sources of metadata to find the data source related to the research question, and bind the data source to the specific research question; in this way, the relevance of the source data to the research question can be measured, and a summary report of the data collection scope and the corresponding relevance can be generated; these reports will be provided to relevant parties to meet research needs without exposing sensitive patient data. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0030] The present invention is further described in detail below in conjunction with embodiments. The following embodiments are for explanation of the present invention but the present invention is not limited to the following embodiments.

[0031] Specific embodiment 1: Please refer to Figure 1 A method for classifying, tracing and recommending data sources for clinical research data comprises the following steps:

[0032] S1: Access to clinical study case report form questions;

[0033] In step S1, the case report form is an electronic medical record report form; preferably, the format of the electronic medical record report form is a table format, such as: Excel table format; the electronic medical record report form includes multiple worksheets, each worksheet records in detail the specific purpose of data collection, such as: collection of drug usage or demographic information; each worksheet has data fields, for example: in the worksheet of drug usage, the data fields include drug name, frequency of use and time of use.

[0034] S2: Recommended categories of relevant hospital source data for answering questions about pathology reports; specifically, they include:

[0035] S201: converting the table name and corresponding data fields of the electronic medical record report form obtained in S1 into metadata text description;

[0036] S202: Use semantic embedding technology and HNSW vector library retrieval technology to classify metadata text descriptions;

[0037] By using the Embedding technology to embed metadata into semantic vector representation, the HNSW algorithm is used to build a database; the semantic similarity retrieval technology is used to quickly match data fields with metadata and their data sources: metadata and data sources related to data fields can be quickly found based on specific data fields;

[0038] Classify metadata text descriptions into several different categories, such as: basic patient information, drug prescriptions, laboratory tests, ward rounds reports, past and present medical history, surgical history, vital signs, etc.

[0039] S203: Using the combination of Embedding technology and HNSW algorithm, a data world model is constructed to process unstructured and structured metadata respectively;

[0040] The unstructured metadata is structured into a schema with patient ID, visit ID, recording time, source data name, text content, additional field name, and additional field results; other content stored in the hospital data system should be linked and sent to the additional field results, and metadata corresponding to the additional content should be added to the additional field name to describe the results; the structured metadata is structured into the following schema: patient ID, visit ID, recording time, source data name, specified data field name, specified field results, additional field name, additional field results; for structured data fields, common data fields are used as specified fields, and additional fields should cover other data stored in the hospital data system.

[0041] Source data name, for example: laboratory test;

[0042] Specify data field names, such as laboratory items, results, upper and lower limits of normal results;

[0043] The specified field results, for example: hbA1c, 7.0, 6.0, 7.2, correspond to the laboratory items, results, and upper and lower limits of normal results in the specified data field names.

[0044] An example of a single piece of metadata that can be stored in a vector database after being processed by this framework:

[0045]

[0046] The source data names in the unstructured metadata include discharge records, ward rounds records, etc.

[0047] Additional field names in the unstructured metadata include doctor name, change status, etc.

[0048] The specified data field names include laboratory test name, test result, test result lower limit, test result upper limit, etc.

[0049] S3: Generate ETL data transmission program and code for transmitting source data; specifically, it includes:

[0050] S301: Apply the algorithm of S2 to pre-process the unstructured metadata and / or structured metadata while retaining the corresponding data sources; for example, mapping the drug order metadata with the data sources such as the doctor's order, ward round record, and discharge record;

[0051] S302: Generate a data mapping document, which contains sample SQL (Structured Query Language) codes required for each type of metadata and a description of the use of metadata in the case report form.

[0052] S4: Obtain anonymized sample source data;

[0053] The data mapping document in step S302 is desensitized to obtain anonymized sample source data after desensitization.

[0054] S5: Search and sort source data categories for each research question;

[0055] Traverse each research question in the case report form, and use the technology of S202 to match the N text segments in the sample library that are most relevant to the research question and the data sources corresponding to the text segments; then count the frequency of occurrence of each data source and sort them in descending order according to the frequency. The present invention will retain the top three data sources with the highest evaluation rate as the recommended data sources for the research question;

[0056] The N value can be customized, and preferably, it is set to 10 in the present invention.

[0057] Preferably, the research question is a research field in a case report form, such as whether there is diabetes, hypertension, coronary heart disease, etc.

[0058] S6: Manually determine the traceability path;

[0059] A manual review module is provided, which is used to review and select data sources related to each research question in the case report form;

[0060] Specifically, based on the recommended data sources for each field obtained in S5, a manual review module is added; the manual review module is designed to review and select the data sources most relevant to each research question in the pathology report form from the system recommended data sources, avoid data source tracing errors caused by system errors, and achieve a better balance between the efficiency and accuracy of the system.

[0061] S7: Generate data interaction specification form;

[0062] The result form in S302 is updated using the data source after manual review in S6 as the final data interaction standard form.

[0063] The cases updated in each system can be updated regularly or manually when there is a temporary need.

[0064] Using the above method, you can perform the following operations to obtain data:

[0065] 1. Define data structure: determine the format required for in-hospital source data and research fields when transmitting data;

[0066] 2. Use historical desensitized data to build a data model;

[0067] 3. According to the specific research fields in the clinical research case report form (such as whether the patient has coronary heart disease), match the data sources of the top K possibilities;

[0068] 4. Manually review and retain only the most relevant data sources and their metadata;

[0069] 5. Generate data interaction forms for specific research fields (such as whether the patient has coronary heart disease).

[0070] The above is only a preferred specific implementation method of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes within the technical scope disclosed in the present application according to the technical solution and improved ideas of the present application, which should be covered by the protection scope of the present application.

Claims

1. A method for data source classification, tracing and recommendation of clinical research data, characterized by: The following steps are involved: S1: Access to clinical study case report form questions; S2: Recommended categories of relevant hospital source data for answering questions about pathology reports, including: S201: converting the table name and corresponding data fields of the electronic medical record report form obtained in S1 into metadata text description; S202: Use semantic embedding technology and HNSW vector library retrieval technology to classify metadata text descriptions; embed metadata into semantic vector representation using Embedding technology, and then use HNSW algorithm ideas to build a database; use semantic similarity retrieval technology to achieve rapid matching of data fields with metadata and their data sources; S203: Using the Embedding technology and HNSW algorithm, a data world model is constructed to process unstructured and structured metadata respectively; S3: Generate ETL data transmission program and code for transmitting source data; specifically, it includes: S301: applying the algorithm of S2 to pre-process the unstructured metadata and / or structured metadata while retaining the corresponding data source; S302: Generate a data mapping document, which includes sample SQL codes required for each type of metadata and a description of the use of metadata in the case report form; S4: Obtain anonymized sample source data; S5: Search and sort source data categories for each research question; S6: Manually determine the traceability path; S7: Generate a data interaction specification form.

2. A method for data source classification, tracing and recommendation of clinical research data according to claim 1, characterized in that: In step S203, the unstructured metadata is structured into a schema having patient ID, visit ID, recording time, source data name, text content, additional field name, and additional field result; Retrieval of other content stored in the hospital data system should be linked and sent under the additional field results, and metadata corresponding to the additional content should be added to the additional field name to describe the result.

3. A method for data source classification, tracing and recommendation of clinical research data according to claim 1, characterized in that: In step S203, the structured metadata is structured into the following schema: patient ID, visit ID, recording time, source data name, specified data field name, specified field result, additional field name, additional field result; For structured data fields, common data fields are taken as designated fields, while additional fields cover other data stored in the hospital data system.

4. A method for data source classification, tracing and recommendation of clinical research data according to claim 1, characterized in that: In step S5, each research question in the case report form is traversed, and the N text segments in the sample library that are most relevant to the research question and the data sources corresponding to the text segments are matched using the technology in S202; then the frequency of occurrence of each data source is counted and sorted in descending order according to frequency, and the top three data sources with the highest evaluation rate are retained as the recommended data sources for the research question; The research question is the research field in the case report form.

5. A method for data source classification, tracing and recommendation of clinical research data according to claim 4, characterized in that: N is a custom value.

6. A method for data source classification, tracing and recommendation of clinical research data according to claim 4, characterized in that: The value of N is 10.

7. A method for data source classification, tracing and recommendation of clinical research data according to claim 1, characterized in that: In step S6, a manual review module is provided, and the manual review module is used to review and select data sources related to each research question in the case report form.

8. The method for data source classification, tracing and recommendation of clinical research data according to claim 1, characterized in that: In step S7, the result form in S302 is updated using the data source after manual review in S6 as the final data interaction standard form.

9. A method for data source classification, tracing and recommendation of clinical research data according to claim 1, characterized in that: In step S4, the data mapping document in S302 is desensitized to obtain anonymized sample source data after desensitization.

10. The method for data source classification, tracing and recommendation of clinical research data according to claim 1, characterized in that: Each metadata has a corresponding data field structure.

Images