Method and system for carrying out standard labeling on case report form

Abstract

The invention discloses a method and a system for standard labeling of a case report form, and the method comprises the following steps: S1, analyzing information in the case report form, and matching the information with information in a research design specification; s2, matching the research design specification with the standardized data set; s3, according to the matching relationship between the case report form and the research design specification and the matching relationship between the research design specification and the standardized data set, obtaining and recording the mapping relationship between the case report form and the standardized data set; and S4, generating a standardized label for the case report form. According to the scheme, the SDS file existing in each test item is used as the intermediate file between the case report form and the standardized data set to achieve automatic labeling, additional variable label adding actions are not needed, and a user only needs to upload the case report form needing to be labeled and the corresponding SDS file to a system to achieve automatic labeling. And the labeling accuracy is ensured while the labeling efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of clinical trials, and in particular relates to a method and system for standardizing case report forms. Background Art

[0002] The annotated case report form (aCRF) of a drug clinical trial is one of the important materials submitted by the sponsor to the regulatory agency. The submission of aCRF ​​is to record the data and observations of patients in the clinical trial for subsequent analysis and review by the regulatory agency. According to the "Guidelines for Submission of Drug Clinical Trial Data (Trial)" issued by the Center for Drug Evaluation of the State Drug Administration, the sponsor needs to submit annotated case report form data to the regulatory agency. Researchers usually need to collect, organize, and analyze annotated case report form data in a standardized manner and present these data clearly, including a specific description of the mapping relationship between the information units (i.e., field information) in the collected subject data (electronic case report form eCRF or paper case report form CRF) and the corresponding variables or variable values ​​in the submitted original data set, so as to facilitate the review of the regulatory agency. When submitting data to different regulatory agencies, it is necessary to follow the standards of each agency. For example, when submitting data to the FDA, it is necessary to refer to the FDA's technical consistency guidelines for research data to meet the FDA's submission requirements for aCRF; when submitting data to PMDA, it is necessary to refer to PMDA's technical consistency guidelines for research data to meet PMDA's submission requirements for aCRF.

[0003] At present, the annotation criteria of case report forms in the clinical industry are mostly implemented based on the standards of the Association for Clinical Data Exchange Standards. However, in the traditional annotation of case report forms (aCRF), most of the adjustments to the standardized annotation format and content need to be done manually. Manual annotation is time-consuming and labor-intensive, inefficient, and manual operation is prone to non-standard annotation. In addition, in traditional annotation work, the process of annotating case report forms, writing programming instructions, and data standardization programming is repeated.

[0004] To this end, people have conducted a lot of research. For example, a Chinese patent discloses a method and device for multi-data source acquisition and conversion in clinical trials [publication number: CN111916161B]. The key process of this method is that the test data contains variables and variable labels, and the variable labels are composed of strings, which correspond to the target strings in the fuzzy matching algorithm; the standardized SDTM database is provided with SDTM domains and each domain variable and its SDTM controlled terms, and the domain variables and SDTM controlled terms are composed of strings, which correspond to the pattern strings in the fuzzy matching algorithm. The pattern string set is converted into a tree-like finite state automaton; the last character of the shortest pattern string in the string tree composed of the tree-like finite state automaton is aligned with the last character of the target string; the string tree is compared from front to back with the characters in the target string, and when there is a mismatch, the forward jump length of the string tree according to the bad character jump method and the forward jump length of the string tree according to the good prefix jump method are calculated, and the larger jump length is taken for jumping; if any pattern string completely matches at least part of the continuous string in the target string, the pattern string is judged to match the target string.

[0005] The above method can replace the original manual conversion process with an intelligent conversion process, but there are still defects, such as the need to manually add the variable labels of the test data to complete the fuzzy matching. In addition, due to the standardization of the labels, the string of the variable label may not match the string of the controlled term of SDTM. For example, the test data contains the variable "gender" and its variable label "Gender", while the domain variable corresponding to SDTM is "SEX" and the controlled terms are "M" and "F". At this time, this method cannot complete the matching of this situation. Summary of the invention

[0006] The purpose of the present invention is to address the above-mentioned problems and to propose a method for standardizing case report forms;

[0007] Another object of the present invention is to provide a system for standardizing the annotation of case report forms.

[0008] A method for standardizing a case report form, the method comprising:

[0009] S1. Parse the information in the case report form and match it with the information in the study design specifications;

[0010] S2. Match study design specifications with standardized datasets;

[0011] S3. According to the matching relationship between the case report form and the research design specification and the matching relationship between the research design specification and the standardized data set, obtain and record the mapping relationship between the case report form and the standardized data set;

[0012] S4. Generate standardized annotations for case report forms.

[0013] In the above method for standardizing case report forms, each trial project includes a study design specification and at least one case report form;

[0014] The research design specifications are used to describe the design and implementation details of the corresponding clinical trials, including the structure of the case report form, the definition, structure, and attributes of the data fields, and the rules and standard specifications for data collection.

[0015] In the above-mentioned method for standardizing case report forms, in step S1, the fields of the blank case report form are read, and according to the research design specification, the read fields are accurately matched with the information in the research design specification. SDS defines the data structure and attributes of each field in CRF. CRF needs to collect data strictly in accordance with the design of SDS. Therefore, under normal circumstances, all contents in CRF should have corresponding information in SDS. Therefore, for the same test project, CRF can achieve an accurate match with SDS.

[0016] In the above-mentioned method for standardizing the case report form, in step S2, natural language processing and machine learning methods are used to match the research design specifications with the standardized data set.

[0017] In the above method for standardizing case report forms, step S2 specifically includes:

[0018] Process the form in the research design specification to obtain form features;

[0019] Process the domains in the standardized dataset to obtain domain features;

[0020] Obtaining a matching relationship between the form and the domain based on the form features and the domain features;

[0021] Process the fields in the form to obtain form field features;

[0022] Processing the fields in the domain to obtain domain field features;

[0023] Based on the domain field characteristics, the form field characteristics, and the matching relationship between the form and the domain, the matching relationship between the form field and the domain field is obtained.

[0024] For some fields, there are also matching relationships between field values, that is, there are three levels of annotation: form, field, and term. If a field has a drop-down box for selection, such as gender, and the field value has two possible values: male / female, then it has an annotation at the field value level. If a field is used to fill in information, such as adverse reactions, then it does not need to be annotated at the field value level.

[0025] The matching relationship between form and domain and the matching relationship between form field and domain field can be output together using the model. In the model, the matching relationship between form and domain is obtained first, then the matching relationship between form field and domain field is obtained, and finally they are output together. It can also be output successively, first the matching relationship between form and domain is output, and then the former is used together with the domain field features and the form field features as input to obtain the matching relationship between form field and domain field.

[0026] In the above method for standardizing the case report form, the labels of the form in the research design specification are characterized using TF-IDF (term frequency-inverse document frequency) to obtain form label features, and the domain labels in the standardized data set are processed to obtain domain label features; the form features include the form label features, and the domain features include the domain label features;

[0027] Use word embedding (such as Word2Vec), one-hot encoding, and TF-IDF methods to process form fields and domain fields to obtain form field features and domain field features respectively;

[0028] The trained multi-layer perceptron is used to output the matching relationship between the form and the domain, and the matching relationship between the form field and the domain field based on the form label features, the domain label features, the form field features, and the domain field features.

[0029] In the above-mentioned method for standard annotation of case report forms, the one-hot encoding (OHE), word embedding (such as Word2Vec), word statistics method (TF-IDF) and multilayer perceptron (MLP) constitute a annotation model. During the annotation model training process, word embedding (including Word2Vec) and multilayer perceptron are optimized through feature engineering;

[0030] The dataset for training the annotation model is constructed as follows:

[0031] Prepare study design specifications and annotated case report forms for multiple different trial projects;

[0032] Obtaining a mapping relationship between a case report form and a standardized data set according to the annotations of the annotated case report form;

[0033] Parse information from case report forms and match it with information from study design specifications;

[0034] Label samples according to the mapping relationship and matching results for the research design specifications;

[0035] At this point, a training dataset consisting of a standardized dataset, research design specifications, and sample labels is obtained.

[0036] In the above method for standardizing the case report form, the method further includes:

[0037] In response to a labeling request of a case report form, partitioning the current interface;

[0038] Displaying the PDF page of the read case report form in a partition interface;

[0039] At the same time, a marking area is provided in another partition interface;

[0040] According to the mapping relationship between the acquired case report form and the standardized data set, pre-fill the annotation information in the annotation area;

[0041] According to the annotation information filled in the annotation area, add the corresponding annotation box at the corresponding position of the PDF page.

[0042] In the above method for standardizing the case report form, the method further includes:

[0043] According to the standardized annotation results of the case report form, the SDTM mapping relationship file is exported.

[0044] A system for standardizing case report forms, wherein annotated case reports are generated by the method described.

[0045] The advantages of the present invention are:

[0046] 1. This solution uses an intermediate file between the case report form and the standardized data set to achieve automatic annotation, which can ensure the accuracy and effectiveness of the annotation, and ensure the accuracy of the annotation while improving the annotation efficiency;

[0047] 2. This solution uses the existing SDS file of each test project as the intermediate file between the case report form and the standardized data set to achieve automatic labeling. No additional variable label adding action is required, which simplifies the automatic labeling workflow. Users only need to upload the case report form and the corresponding SDS file to be labeled to the system, and the system can automatically label and display the labeled case report form and label column in a visual way, which is convenient for users to check and modify;

[0048] 3. This solution uses the characteristics of SDS files containing the data structure and attributes of each field in the case report form to accurately match the information in the case report form with the SDS information. At the same time, it uses the rich information in SDS, such as data type, length, description and other metadata that cannot be reflected in the case report form, to use machine learning and natural language processing to learn the matching relationship between the forms, fields, etc. in SDS and the standardized data set based on the rich information in SDS, and then match the case report form with the standardized data set based on the previous precise matching results, which can achieve more accurate automatic labeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flow chart of a method for standardizing a case report form according to the present invention;

[0050] Figure 2 The present invention provides the implementation principle of standard marking on case report forms;

[0051] Figure 3 The process of using the system for standardizing case report forms of the present invention;

[0052] Figure 4 A diagram of the reasoning principle of the annotation model in the method for performing standard annotation on case report forms of the present invention;

[0053] Figure 5 This is a user interface diagram of a system for standardizing case report forms in the present invention. DETAILED DESCRIPTION

[0054] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] This protocol provides a method for standardizing case report forms, such as Figure 1 and Figure 2 As shown, the method includes:

[0056] S1. Parse the information in the case report form (CRF) and match it with the information in the study design specification (SDS); specifically, extract the string on the case report form in PDF format, and then process it through the following process to match the content in the SDS:

[0057] Ignore line breaks, multiple consecutive spaces and other placeholders from the case report form string and convert them all into single spaces;

[0058] Change all symbols from the case report form string to half-width symbols and all letters to uppercase letters;

[0059] Match with the content in SDS.

[0060] S2. Match the research design specification SDS with the standardized data set. This embodiment takes the standardized data set SDTM as an example for explanation. In actual use, it is not limited to SDTM.

[0061] S3. According to the matching relationship between the case report form CRF and the research design specification SDS and the matching relationship between the research design specification SDS and the standardized data set SDTM, obtain and record the mapping relationship between the case report form CRF and the standardized data set SDTM;

[0062] S4. Generate standardized annotations for the case report form CRF to obtain the annotated case report form aCRF, and derive the mapping relationship.

[0063] Each trial project will include a study design specification SDS and at least one case report form CRF. The study design specification SDS describes the design of the clinical study and the data collection method, including the structure of the CRF, the definition of data fields, the rules and standards for data collection, etc. It is a database specification exported from the EDC system.

[0064] In step S1, the fields of the blank case report form CRF are read, and according to the research design specification SDS, the read fields are accurately matched with the information in the research design specification SDS. SDS defines the data structure and attributes of each field in CRF, and CRF needs to collect data strictly according to the design of SDS. Therefore, under normal circumstances, all contents in CRF should have corresponding information in SDS. Therefore, for the same test project, CRF can achieve an accurate match with SDS.

[0065] like Figure 3 As shown, when it is put into use, the user only needs to import the research design specification file SDS of the corresponding test project and the blank case report form CRF that needs to be labeled into this system. The system automatically labels the blank case report form CRF based on the research design specification SDS and the blank case report form CRF that needs to be labeled to obtain the annotated case report form aCRF, and can export the mapping relationship between the case report form and the standardized data set - SDTM mapping file, which is used for subsequent programming manual writing.

[0066] Furthermore, this embodiment uses natural language processing and machine learning methods to match the research design specification SDS with the standardized data set SDTM. Figure 4 As shown, TF-IDF (term frequency-inverse document frequency) is used to characterize the labels of the form in the research design specification (SDS) to obtain the form label features, and the domain labels in the standardized data set (SDTM) are processed to obtain the domain label features;

[0067] Use Word2Vec, One-Hot Encoding (OHE) and TF-IDF (term frequency-inverse document frequency) methods to process form fields and domain fields, and obtain word vectors of form field features and domain field features respectively;

[0068] A multi-layer perceptron (MLP) is used to obtain the matching relationship between the form and the domain, and the matching relationship between the form field and the domain field based on the form label features, domain label features, form field features, and domain field features.

[0069] The above-mentioned One-Hot Encoding (OHE), Word2Vec, TF-IDF (Term Frequency - Inverse Document Frequency) and Multilayer Perceptron (MLP) constitute a labeling model. During the labeling model training process, TF-IDF, Word2Vec and Multilayer Perceptron are optimized through feature engineering.

[0070] Specifically, the dataset for training the annotation model is constructed as follows:

[0071] Prepare study design specification SDS and annotated case report forms (aCRF) for multiple different trial projects;

[0072] According to the annotations of the annotated case report form aCRF, the mapping relationship between the case report form and the standardized data set SDTM is obtained;

[0073] Parse the information in the case report form aCRF ​​and match it with the information in the study design specification SDS;

[0074] According to the mapping relationship between aCRF ​​and SDTM and the matching results between aCRF ​​and SDS, sample labels are added to the research design specification SDS, that is, the research design specification SDS with sample labels for some data is obtained. For example, the VSDAT in the SDS (the "VSDAT" field in the "vital signs record" form in the SDS) is added with a sample label - VS.VSDTC (the "VSDTC" field in the "VS" domain in the SDTM).

[0075] At this point, the training dataset consisting of the standardized dataset SDTM—research design specification SDS—sample labels is obtained.

[0076] The training data set is then used to train the annotation model. The trained annotation model can output the matching relationship between each data specification in the research design specification SDS and the standardized data set SDTM based on the input standardized data set SDTM and the research design specification SDS.

[0077] During the reasoning process, the information in the case report form CRF is first parsed and accurately matched with the information in the study design specification SDS, and the data information matching the information in the case report form CRF is extracted from the study design specification SDS. Then, the aforementioned matching data information, the study design specification SDS, and the standardized data set SDTM are input into the annotation model. The annotation model at least outputs the matching relationship between the data information in the study design specification SDS that matches the case report form CRF information and the standardized data set SDTM.

[0078] Finally, the mapping relationship between the case report form and the standardized data set is obtained based on the matching relationship between the data information and the standardized data set, as well as the matching relationship between the case report form and the research design specifications.

[0079] In order to better illustrate the method of this solution, an example is used to illustrate the process below.

[0080] Suppose a clinical trial is underway to evaluate the safety and effectiveness of a new drug for treating high blood pressure.

[0081] In this trial, one case report form included the following pages:

[0082] Demography

[0083] Adverse Events

[0084] Vital signs

[0085] Drug exposure

[0086] The Demographics page includes the following fields: Subject ID:

[0087] age:

[0088] gender:

[0089] If female, whether you are fertile: Ethnicity:

[0090] The Adverse Events page includes the following fields: Adverse Event Name:

[0091] start date:

[0092] End Date:

[0093] Severity:

[0094] The Vital Signs page includes the following fields: Measurement Date Time:

[0095] body temperature:

[0096] Systolic blood pressure:

[0097] Diastolic blood pressure:

[0098] Result judgment:

[0099] If the result is abnormal, is it clinically significant? The Drug Exposure page contains the following fields: Date of Administration:

[0100] Route of administration:

[0101] Planned dose:

[0102] Actual dosage:

[0103] Whether fasting:

[0104] The research design specification SDS is prepared to contain the following data information tables: Page Basic Information

[0105]

[0106] Demographic records

[0107]

[0108] Adverse event records

[0109] Field Name Field Labels Field number Input Type Field length Field Format AETERM Adverse event name: 1 Long text 200 AESTDAT start date: 2 Date box 20 yyyy-mm-dd AEENDAT End Date: 3 Date box 20 yyyy-mm-dd AESEV Severity: 4 Vertical radio button 20

[0110] Vital signs recording

[0111]

[0112] Drug exposure records

[0113] Field Name Field Labels Field number Input Type Field length Field Format EXDAT Date of administration: 1 Date box 20 yyyy-mm-dd ROUTE Route of administration: 2 Text Box 20 EXPDOSE Planned dose: 3 Text Box 20 EXPDOSEU Planned dosage unit: 4 Text Box Presets 20 mg EXDOSE Actual dosage: 5 Text Box 20 EXDOSEU Actual dosage unit: 6 Text Box Presets 20 mg EXLIM Whether fasting: 7 Horizontal radio button 2

[0114] …

[0115] The above pilot projects may involve the following domains of the standardized dataset SDTM:

[0116] DM (Demographics)-demographic data, including basic information of the subjects such as age, gender, race, ethnicity, etc.

[0117] AE (Adverse Events) - adverse events, records the adverse events that occur to the subjects, including event name, start time, end time, severity, etc.

[0118] VS (Vital Signs)-vital signs data, including weight, height, systolic blood pressure, diastolic blood pressure, heart rate, body temperature, etc. It is a key endpoint indicator data set for measuring drug effectiveness.

[0119] EX (Exposure)-drug exposure information, including the administration time, planned dosage, actual dosage, administration method, and administration frequency of the trial drug.

[0120] CM (Concomitant Medications)-concomitant medication data, recording all other drugs used by the subjects during the trial.

[0121] LB (Laboratory Test Results) - Laboratory test results, including blood routine, blood biochemistry, urine routine and other indicators, used for drug safety assessment.

[0122] MH (Medical History)-Medical History, records the subject's diagnosis history of hypertension and other relevant medical history.

[0123] SV (Subject Visits) - records the subject's visit information, including visit name, planned visit time, whether the visit was missed, etc.

[0124] DS (Disposition) - Subject distribution, records whether each milestone event of the subject in the clinical trial occurs and when it occurs.

[0125] …

[0126] Each domain contains at least one field. For example, the DM domain contains the following fields:

[0127]

[0128]

[0129] …

[0130] After importing the blank case report form CRF and the research design specification SDS into this system, the system will match the demographic records in the CRF with the demographic records in the SDS, and match the subject number, age, age unit, gender, fertility, ethnicity, etc. to the corresponding fields in the SDS; match the adverse events in the CRF with the adverse events in the SDS, and match the adverse event name, start date, end date, severity, etc. to the corresponding fields in the SDS; ...in this way, the information matching between CRF and SDS is completed.

[0131] Then, the trained annotation model is used to match the study design specification SDS with SDTM, such as matching demographic records to DM, vital signs information to VS, adverse events to AE, drug exposure information to EX, and vital signs, drug exposure and other visit information to SV.

[0132] According to the matching relationship between CRF and SDS, and the matching relationship between SDS and SDTM, the system will obtain the mapping relationship between CRF and SDTM. For example, the gender in the demographic record is mapped to SEX in SDTM.DM, the age is mapped to AGE in SDTM.DM, and "if female, whether fertile" is matched to SUPPDM.CHILDPOT; in the vital signs record, the measurement date is matched to VS.VSDTC, and the temperature, systolic blood pressure, diastolic blood pressure results and result judgments are matched to VS.VSORRES with VS.VSTESTCD of "TEMP", "SYSBP", "DIABP" and "INTP" respectively. The clinical significance of abnormal results is matched to SUPPVS.VSCLSP...

[0133] In this way, each specific field in CRF can be correctly mapped to SDTM, thereby achieving automatic labeling of data.

[0134] Finally, the required SDTM mapping relationship file is automatically generated based on the two matching results. This file provides the mapping relationship between the fields in the CRF and the domains and variables of the SDTM, which can be exported by the user.

[0135] Furthermore, if Figure 5 As shown, the annotation process of this method performs visual automatic annotation, including:

[0136] In response to a labeling request of a case report form, partitioning the current interface;

[0137] The PDF page of the read case report form is displayed in a partition interface, that is, Figure 5 2 areas in it.

[0138] At the same time, a marking area is provided in another partition interface, namely Figure 5 3 areas in; Figure 5 The 1 area in the figure is the basic information of the current aCRF.

[0139] According to the mapping relationship between the obtained case report form and the standardized data set, pre-fill the annotation information in the annotation area - 3 area, such as Figure 5 The DM domain and DS domain are pre-filled in the CRF, and the domain description is automatically filled in according to the pre-filled domain results. Users can intuitively see the matching situation and can modify it manually. Based on the analysis of the PDF version of the CRF on the left, the system automatically adds the matching domains, fields, and values ​​on the right to the appropriate position of the CRF on the left in the form of a annotation box. Users can manually drag the annotation box to change the position. Through this visualization method, users can intuitively see the process and results of the system's automatic annotation and manually modify the automatic annotation process and results to improve the adaptability and accuracy of the system.

[0140] This scheme takes advantage of the uniformity of information between SDS and CRF, as well as the richness of SDS information, and uses SDS as the intermediate file for mapping between CRF and SDTM, and accurately matches the CRF information that needs to be labeled to SDS. Due to the characteristics of the experimental design, SDS and CRF can be fully matched, but there will be some difficulties between SDS and SDTM. In this regard, this scheme proposes to use natural language and machine learning to automatically learn the matching relationship between SDS tables, fields, values ​​and SDTM based on features. The training data can use a variety of labeled aCRFs to achieve labeling of key data in the corresponding SDS. Due to the rich data form of SDS, when the training data set composed of standardized data sets-research design specifications-sample labels is used to train the model, better training results can be obtained, thereby meeting the needs of automatic matching.

[0141] The specific embodiments described in this embodiment are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the attached claims.

[0142] Although the terms such as case report form, research design specification, standardized data set, CRF, aCRF, SDTM, SDS are used more frequently in this article, the possibility of using other terms is not excluded. The use of these terms is only to more conveniently describe and explain the essence of the present invention; interpreting them as any additional limitation is contrary to the spirit of the present invention.

Claims

1. A method for standardizing a case report form, characterized in that: The method includes: S1. Parse the information in the case report form and match it with the information in the study design specifications; S2. Match study design specifications with standardized datasets; S3. According to the matching relationship between the case report form and the research design specification and the matching relationship between the research design specification and the standardized data set, obtain and record the mapping relationship between the case report form and the standardized data set; S4. Generate standardized annotations for case report forms.

2. The method for standardizing case report forms according to claim 1, characterized in that: Each trial project includes a study design specification and at least one case report form; The research design specifications are used to describe the design and implementation details of the corresponding clinical trials, including the structure of the case report form, the definition, structure, and attributes of the data fields, and the rules and standard specifications for data collection.

3. The method for standardizing case report forms according to claim 2, characterized in that: In step S1, the fields of the blank case report form are read, and according to the research design specification, the read fields are accurately matched with the information in the research design specification.

4. The method for standardizing case report forms according to claim 2, characterized in that: In step S2, natural language processing and machine learning methods are used to match the study design specifications with the standardized dataset.

5. The method for standardizing case report forms according to claim 4, characterized in that: Step S2 specifically includes: Process the form in the research design specification to obtain form features; Process the domains in the standardized dataset to obtain domain features; Obtaining a matching relationship between the form and the domain based on the form features and the domain features; Process the fields in the form to obtain form field features; Processing the fields in the domain to obtain domain field features; Based on the domain field characteristics, the form field characteristics, and the matching relationship between the form and the domain, the matching relationship between the form field and the domain field is obtained.

6. The method for standardizing case report forms according to claim 5, characterized in that: Using a word statistics method to characterize the notes in the form in the research design specification to obtain form label features, and processing the domain labels in the standardized data set to obtain domain label features, the form features include the form label features, and the domain features include the domain label features; Use word embedding, one-hot encoding and word statistics methods to process form fields and domain fields to obtain form field features and domain field features respectively; A multi-layer perceptron is used to obtain the matching relationship between the form and the domain, and the matching relationship between the form field and the domain field based on the form label features, domain label features, form field features, and domain field features.

7. The method for standardizing case report forms according to claim 1, characterized in that: The one-hot encoding, word embedding, word statistics method and multi-layer perceptron constitute a labeling model. During the labeling model training process, word embedding and multi-layer perceptron are optimized in parameters through feature engineering. The dataset for training the annotation model is constructed as follows: Prepare study design specifications and annotated case report forms for multiple different trial projects; Obtaining a mapping relationship between a case report form and a standardized data set according to the annotations of the annotated case report form; Parse information from case report forms and match it with information from study design specifications; Label samples according to the mapping relationship and matching results for the research design specifications; At this point, a training dataset consisting of a standardized dataset, research design specifications, and sample labels is obtained.

8. The method for standardizing case report forms according to claim 7, characterized in that: The method further comprises: In response to a labeling request of a case report form, partitioning the current interface; Displaying the PDF page of the read case report form in a partition interface; At the same time, a marking area is provided in another partition interface; According to the mapping relationship between the acquired case report form and the standardized data set, pre-fill the annotation information in the annotation area; According to the annotation information filled in the annotation area, add the corresponding annotation box at the corresponding position of the PDF page.

9. The method for standardizing case report forms according to claim 8, characterized in that: The method further comprises: According to the standardized annotation results of the case report form, the SDTM mapping relationship file is exported.

10. A system for standardizing case report forms, characterized in that: An annotated case report is generated by the method of any one of claims 1-9.

Images