Clinical trial data risk early warning system based on big data analysis

Abstract

The present application provides a clinical trial data risk warning system based on big data analysis, which involves the field of risk warning technology, including: receiving multi-source clinical trial data from multiple research centers; obtaining multi-source background information; adaptively optimizing multiple initial cleaning rule generation strategies; performing local specificity identification on multi-source clinical trial data, standardizing non-local specific data, and standardizing local specific data; performing anomaly detection and early warning analysis; identifying dynamic changes in real-time risk warning reports, and adjusting the warning level in real time. This application can solve the technical problems in the prior art that the significant differences in data formats, recording standards, and data quality of multiple research centers affect the processing and analysis of data, achieve the technical goal of differentiated processing of multi-source heterogeneous clinical trial data, and achieve the technical effect of reducing errors and optimizing the accuracy of data analysis.

Description

Technical Field

[0001] The present application relates to the field of risk warning technology, and in particular to a clinical trial data risk warning system based on big data analysis. Background Art

[0002] In modern clinical trials, data management and analysis are key links to ensure the validity and reliability of trial results. In current technologies, the processing of clinical trial data mainly relies on data collected from multiple research centers, covering background information of subjects, laboratory results, drug reactions, etc. These data come from various sources, including electronic medical record systems of medical institutions, laboratory test results, and drug use records. Although existing technologies have made significant progress in data integration and analysis, they still face some key shortcomings.

[0003] At present, in the existing technology, since clinical trials usually involve multiple research centers, there may be significant differences in data formats, recording standards and data quality between different centers. This heterogeneity makes it extremely complicated to quickly process and analyze data from different sources. For example, a research center may use different laboratory testing methods and units, resulting in errors in data integration. In addition, the diversity of the subjects' background information, such as age, gender, geographic location, etc., may also affect the accuracy of data analysis.

[0004] In summary, the existing technology has the problem of heterogeneity due to the involvement of multiple research centers in clinical trials, resulting in significant differences in data formats, recording standards and data quality among different centers, making it extremely complicated to quickly process and analyze data from different sources. This data heterogeneity not only causes errors in the data integration process, but also may affect the accuracy of the analysis of the diversity of the subject's background information, further affecting the accuracy and reliability of the final clinical trial results. Summary of the invention

[0005] The purpose of this application is to provide a clinical trial data risk warning system based on big data analysis to solve the technical problems in the prior art that clinical trials involve multiple research centers, resulting in significant differences in data formats, recording standards and data quality, which further affects the rapid processing of data and the complexity of analysis, leading to errors in data integration and inaccurate analysis results.

[0006] In view of the above problems, the present application provides a clinical trial data risk early warning system based on big data analysis, including: a data import module, used to receive multi-source clinical trial data of multiple subjects imported by multiple research centers; a background information acquisition module, used to obtain multi-source background information of the multiple subjects; a dynamic adaptive cleaning module, used to adaptively optimize multiple initial cleaning rule generation strategies of the multiple research centers to obtain multiple cleaning rule generation strategies; a local specificity identification module, used to perform local specificity identification on the multi-source clinical trial data according to the multi-source background information, obtain local specific data, and standardize the non-local specific data through the non-local cleaning rules obtained based on the cleaning rule generation strategy The method comprises the following steps: a first step is to process the non-local clinical trial data, perform standardization processing on the local specific data by using the local cleaning rules obtained based on the cleaning rule generation strategy, obtain the local clinical trial data, and obtain the clinical trial data by combining the non-local clinical trial data and the local clinical trial data, wherein the non-local clinical trial data is obtained by removing the local specific data from the multi-source clinical trial data; an early warning module is used to perform anomaly detection on the non-local clinical trial data and the local clinical trial data to obtain abnormal data, and perform early warning analysis on the abnormal data to generate a real-time risk early warning report; and a real-time risk adjustment module is used to identify the dynamic changes of the real-time risk early warning report, and adjust the early warning level in real time.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] A data import module is used to receive multi-source clinical trial data of multiple subjects imported by multiple research centers; a background information acquisition module is used to obtain multi-source background information of the multiple subjects; a dynamic adaptive cleaning module is used to adaptively optimize multiple initial cleaning rule generation strategies of the multiple research centers to obtain multiple cleaning rule generation strategies; a local specificity identification module is used to perform local specificity identification on the multi-source clinical trial data according to the multi-source background information to obtain local specific data, perform standardization processing on the non-local specific data through the non-local cleaning rules obtained based on the cleaning rule generation strategy to obtain non-local clinical trial data, perform standardization processing on the local specific data through the local cleaning rules obtained based on the cleaning rule generation strategy to obtain local clinical trial data, combine the non-local clinical trial data and the local clinical trial data, and perform standardization processing on the local specific data through the local cleaning rules obtained based on the cleaning rule generation strategy to obtain local clinical trial data. The clinical trial data is obtained by using clinical trial data from multiple sources, wherein the non-local clinical trial data is obtained by removing the local specific data from the multi-source clinical trial data; an early warning module is used to perform abnormality detection on the non-local clinical trial data and the local clinical trial data to obtain abnormal data, and to perform early warning analysis on the abnormal data to generate a real-time risk early warning report; a real-time risk adjustment module is used to identify dynamic changes in the real-time risk early warning report, and to adjust the early warning level in real time. In other words, the clinical trial data risk early warning system based on big data analysis is a highly integrated clinical trial data processing platform, which includes multiple modules that work together to ensure data accuracy and real-time risk management, and achieves the technical goal of rapid processing of data across multiple research centers, thereby achieving the technical effect of improving data integration accuracy, ensuring accurate analysis results and effectively retaining local specific data.

[0009] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0011] Figure 1This is a schematic diagram of the structure of the clinical trial data risk early warning system based on big data analysis in this application;

[0012] Figure 2 This is a schematic diagram of the process of obtaining multi-source background information of the multiple subjects in the clinical trial data risk warning system based on big data analysis in this application.

[0013] Description of reference numerals:

[0014] Data import module 11, background information acquisition module 12, dynamic adaptive cleaning module 13, local specificity recognition module 14, early warning module 15, real-time risk adjustment module 16. DETAILED DESCRIPTION

[0015] This application provides a clinical trial data risk warning system based on big data analysis, which solves the technical problem in the prior art that clinical trials involve multiple research centers, resulting in significant differences in data formats, recording standards and data quality, further affecting the complexity of rapid data processing and analysis, leading to errors in data integration and inaccurate analysis results. The technical goal of rapid processing of multi-source heterogeneous clinical trial data is achieved, achieving the technical effect of improving the accuracy of data integration, reducing errors and optimizing the accuracy of data analysis.

[0016] Below, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. It should also be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all of them.

[0017] Please refer to the attached Figure 1 , this application provides a clinical trial data risk early warning system based on big data analysis, specifically including:

[0018] The data import module 11 is used to receive multi-source clinical trial data of multiple subjects imported by multiple research centers.

[0019] Specifically, the data import module receives multi-source clinical trial data from multiple research centers. Research centers refer to different medical institutions or scientific research units that are responsible for collecting and submitting data from multiple subjects participating in multi-source clinical trials. Since the data sources and formats of each research center may be different, the data are uniformly collected and processed through the data import module. Multiple subjects refer to the subjects participating in the trial from different geographical locations, backgrounds, age groups or health conditions, so the data differences are significant. For example, one research center may focus on elderly patients with cardiovascular disease, while another center focuses on diabetic patients, and the data differences require the system to effectively integrate and clean up. Multi-source clinical trial data refers to various types of data collected from different subjects, including but not limited to laboratory test results, drug reactions, adverse event records, etc. For example, one research center may upload blood glucose monitoring data, and another research center may upload electrocardiogram data. Therefore, the data import module not only needs to receive this data, but also needs to be able to process the data formats and contents from different sources to ensure that the data is consistent and non-missing in subsequent analysis.

[0020] The background information acquisition module 12 is used to acquire multi-source background information of the multiple subjects.

[0021] Specifically, the background information acquisition module is responsible for integrating the subject background information from different sources to generate non-missing multi-source background information. The multi-source data fusion module can process and coordinate the data fusion from multiple different data sources, ensuring that the subject background data obtained from different research centers, medical institutions or laboratories can be comprehensively processed in the same system, thereby eliminating the differences between the data. Multiple subjects refer to the subject groups participating in the clinical trial, whose data come from different geographical locations, institutions or experimental environments. Multi-source background information may include age, gender, race, geographical location, lifestyle, medical history, etc. Since the multi-source background information comes from different research centers or databases, there may be inconsistencies or differences in format. Background information refers to various basic data related to the individual subjects, which is used to explain and support the analysis and evaluation of clinical trial data. For example, one research center may provide basic demographic data of the subjects, such as gender and age, and another center may provide more detailed medical history information, such as chronic disease records or drug allergy history. Therefore, data from different sources need to be processed and standardized separately. Multi-source background information is the result of fusion through this module, which contains the non-missing background data of all subjects and removes repeated, inconsistent or formatted information. For example, a subject's medical history data may come from multiple hospitals, which are organized into a clear and non-duplicate record through the multi-source data fusion module to ensure that all background information is verified and ready for subsequent analysis of the clinical trial. For example, suppose a clinical trial involves patients from different countries around the world, and some patients' lifestyle data come from a European research center, while other patients' medical history information comes from medical institutions in Asia. The multi-source data fusion module will match, organize and standardize data from these different sources to form non-missing multi-source background information, ensuring consistent analysis of all subjects and supporting further risk warnings.

[0022] The dynamic adaptive cleaning module 13 is used to adaptively optimize the multiple initial cleaning rule generation strategies of the multiple research centers to obtain multiple cleaning rule generation strategies.

[0023] Specifically, the dynamic adaptive cleaning module is used to optimize and adjust different initial cleaning rule generation strategies from multiple research centers to generate standardized cleaning rules that are applicable to the data of each research center. Among them, the dynamic adaptive cleaning module is a module that can automatically adjust the cleaning strategy according to different data characteristics and backgrounds to ensure that the data of each research center can be accurately processed. Multiple research centers refer to different medical institutions or research units participating in clinical trials. Each research center may have formulated different data cleaning rules according to its own needs and standards. For example, some research centers may give priority to the cleaning of outliers in laboratory data, while another research center may focus more on the processing of drug reaction data. Therefore, multiple cleaning rules will show different requirements and processing methods due to differences in research centers. Multiple cleaning rule generation strategies refer to the generation strategies of specific rules and standards used to process subject data in each research center. Different cleaning rule generation strategies may include cleaning rules for the processing of outliers, the filling method of missing data, the screening criteria for specific data items, etc. For example, one research center may use a fixed threshold to judge the abnormality of laboratory data, while another research center may flexibly set the cleaning criteria based on age or geographic location. Adaptive optimization of multiple initial cleaning rule generation strategies refers to dynamically adjusting the parameters or logic of each initial cleaning rule generation strategy according to the data characteristics and cleaning requirements of different research centers in an automated manner, so that it can adapt to different data environments. For example, automatically adjusting the cleaning threshold according to the laboratory data characteristics of a research center, or optimizing the processing rules of abnormal data for subjects of different age groups. The purpose of adaptive optimization is to allow each rule to adapt to various data scenarios without changing the overall structure. The cleaning rule generation strategy is a different cleaning standard after dynamic optimization, which can be applied to the data from each research center to ensure the consistency and accuracy of the data cleaning process. The cleaning rule generation strategy can not only generate cleaning rules to handle differences in data across centers, but also ensure that the cleaned data is consistent and comparable globally.

[0024] The local specificity identification module 14 is used to perform local specificity identification on the multi-source clinical trial data according to the multi-source background information to obtain local specific data, standardize the non-local specific data by using the non-local cleaning rules obtained based on the cleaning rule generation strategy to obtain non-local clinical trial data, standardize the local specific data by using the local cleaning rules obtained based on the cleaning rule generation strategy to obtain local clinical trial data, and obtain clinical trial data by combining the non-local clinical trial data and the local clinical trial data, wherein the non-local clinical trial data is obtained by removing the local specific data from the multi-source clinical trial data.

[0025] Specifically, the local specificity identification module analyzes multi-source clinical trial data based on multi-source background information and identifies the parts with specific differences or special characteristics, namely local specific data. Multi-source background information refers to the subject background data summarized from different data sources (such as multiple research centers), including gender, age, geographic location, medical history, etc., which helps to understand the individual differences between different subjects and ensure that the data processing is targeted.

[0026] After identifying the local specific data, separate the local specific data from the rest of the data and process them separately. First, use the non-local cleaning rules obtained based on the cleaning rule generation strategy to standardize the non-specific data and generate non-local clinical trial data. Non-local cleaning rules are conventional cleaning standards used to process most subject data and are suitable for subject data that do not have significant differences. For example, conventional blood sugar levels, blood pressure and other data are suitable for the background of most subjects and can be processed according to unified standards.

[0027] Next, the local specific data are standardized using local cleaning rules designed specifically for the specific data to generate local clinical trial data. Local cleaning rules are processing standards specifically tailored for a population of subjects with special backgrounds or differences. For example, subjects at high altitudes may have high red blood cell counts, and these data will be retained and processed according to local cleaning rules instead of being treated as outliers.

[0028] Finally, the processed non-local clinical trial data is combined with the local clinical trial data to generate non-missing clinical trial data. It is worth noting that the non-local clinical trial data is obtained by removing the identified local specific data from the multi-source clinical trial data. This mechanism of data splitting and separate processing ensures that different types of data can be cleaned according to the most appropriate rules, which not only ensures the non-missing nature of the data, but also retains local features with clinical significance. Through such refined processing, more accurate and reliable clinical trial data analysis results can be provided.

[0029] The early warning module 15 is used to perform anomaly detection on the non-local clinical trial data and the local clinical trial data to obtain abnormal data, and perform early warning analysis on the abnormal data to generate a real-time risk early warning report.

[0030] Specifically, the early warning module is responsible for anomaly detection of two types of clinical trial data, namely non-local clinical trial data and local clinical trial data, and then identifying abnormal data. Non-local clinical trial data refers to data that does not have specific individual differences and meets universal standards, usually from routine test results of most subjects. For example, basic health data such as blood sugar and blood pressure of most subjects meet universal health standards and are therefore classified as non-local data. Local clinical trial data refers to data with specific individual background or environmental characteristics, such as higher red blood cell counts in subjects at high altitudes, or unique responses of patients with specific medical histories to drugs, which require special processing during analysis.

[0031] Anomaly detection refers to the analysis of non-local clinical trial data and local clinical trial data to identify values ​​that deviate from the normal or expected range, i.e., abnormal data. For example, if the blood sugar levels of most subjects are within the normal range, but the blood sugar of a subject suddenly increases significantly, it is identified as abnormal data.

[0032] When abnormal data is detected, the early warning module will conduct early warning analysis on it. Early warning analysis aims to evaluate whether the abnormal data may represent potential risks or problems. For example, it may be found that some patients have adverse reactions after taking a certain drug, or that a group of patients in a certain environment show consistent abnormal physiological indicators.

[0033] Based on the analysis results, the system generates a real-time risk warning report that records the identified abnormal data and the potential risks it may cause. For example, if a certain patient group shows adverse drug reactions, the report will indicate this risk and recommend further monitoring or adjustment of the treatment plan. Real-time reports help researchers or doctors assist in identifying and responding to emergencies that may arise in clinical trials, thereby improving the safety and effectiveness of clinical trials.

[0034] The real-time risk adjustment module 16 is used to identify dynamic changes in the real-time risk warning report and adjust the warning level in real time.

[0035] Specifically, the real-time risk adjustment module is used to identify risk situations in real-time risk warning reports and adjust the warning level immediately according to the dynamic changes of risks, so as to ensure that it can flexibly respond to changing risk scenarios and timely increase or decrease the warning level, thereby optimizing risk management and decision support. Real-time risk warning reports are reports generated after potential risks are detected, which include analysis and risk assessment of abnormal situations during the trial. For example, in clinical trials, reports may show that some subjects have adverse reactions due to excessive drug doses, and abnormal data and potential risks are listed in detail through warning reports. Dynamic changes refer to the continuous changes in risk factors over time or data changes, which means that risks may increase or decrease with new data input or changes in external conditions. For example, as clinical trials progress, adverse reactions to drugs may gradually escalate from mild dizziness to serious cardiovascular problems. Warning levels are different warning levels set for different degrees of risk, divided into low, medium, and high levels. Low-level warnings may indicate that attention is needed but no immediate action is required; while high-level warnings mean that there is a serious risk and urgent action is required. For example, if abnormal blood pressure is detected in a small number of patients in the early stage, a medium warning may be issued, but if more patients show similar problems, the warning level will be raised to high, indicating the need for immediate intervention. The real-time risk adjustment module can dynamically analyze the severity of the risk based on the risk changes in the warning report. For example, if more and more patients are found to have adverse reactions in clinical trials, the warning level will be raised in real time from medium to high risk, and the trial managers will be notified to take emergency measures to ensure that risk management is more flexible and efficient, and to avoid delays in responding to potential dangers.

[0036] The clinical trial data risk warning system based on big data analysis can achieve the technical goal of rapid processing of data across multiple research centers, and achieve the technical effects of improving the accuracy of data integration, ensuring the accuracy of analysis results and effectively retaining local specific data.

[0037] Further, if Figure 2 As shown, this application also includes:

[0038] Receive the gender information, age information and geographic location information of the multiple subjects to generate primary background information; receive the basic health status information, chronic disease history information, abnormal laboratory test results information and drug metabolism gene information of the multiple subjects to generate secondary background information; combine the primary background information and the secondary background information to obtain the multi-source background information.

[0039] Specifically, multi-dimensional information of multiple subjects is processed hierarchically to generate non-missing multi-source background information. First, basic demographic data related to the subjects, including gender information, age information, and geographic location information, are received and processed to generate primary background information. Primary background information is the basic attributes of the subjects, which can reflect their physiological differences and living environment. For example, subjects of different genders may have differences in drug metabolism, age will also affect the response to drugs, and geographic location may affect the subjects' living habits and health status.

[0040] Next, more detailed health data is received, including basic health information, chronic disease history information, laboratory test abnormality information, and drug metabolism gene information, thereby generating secondary background information that reflects the subject's long-term health status and its potential response to drugs. For example, data from subjects with diabetes or hypertension in clinical trials may show different characteristics, abnormal items in laboratory test results may indicate potential health risks, and drug metabolism gene information determines the subject's ability to metabolize drugs.

[0041] By integrating all primary and secondary background information, comprehensive multi-source background information is ultimately generated, providing a solid foundation for clinical trial data analysis and personalized risk assessment, ensuring that the system can fully and accurately understand the multi-level background information of each subject, thereby achieving more accurate clinical trial data analysis.

[0042] Furthermore, the present application also includes:

[0043] According to the primary background indicators and the secondary background indicators, missing background information of the multi-source background information is identified to obtain missing background information, and the multi-source background information is screened according to the missing background information to obtain non-missing background information; the multi-source background information is associated with the primary background indicators to generate demographic information association; the multi-source background information is multi-combination associated with the secondary background indicators to generate clinical trial history association and disease characteristic-drug response association; the demographic information association, the clinical trial history association and the disease characteristic-drug response association are introduced into the completion generator, and the non-missing background information and the missing background information are complemented by the completion generator to obtain complement background information; a multi-source association analysis result is constructed according to the sample non-missing background information, the sample missing background information and the sample complement background information, and the complement background information is verified based on the multi-source association analysis result. If the complement verification threshold is met, the multi-source background information is obtained by combining the complement background information and the non-missing background information.

[0044] Specifically, the missing background information is identified and generated based on the primary and secondary background indicators. The primary and secondary background indicators correspond to different levels of subject data. For example, the primary background indicators may include basic information such as gender and age, and the secondary background indicators cover medical history and health status. They can identify the missing key information in the data set and ensure that the subsequent completion work can be targeted.

[0045] Next, based on the missing background information, the multi-source background information is screened, incomplete or invalid data is removed, and non-missing background information is generated for subsequent association analysis and completion. For example, if the primary indicator information such as age and gender of the subject is not missing, but the medical history or drug metabolism gene data is missing, the non-missing part is retained and the missing items are marked for subsequent processing.

[0046] Then, we conduct correlation analysis on multi-source background information through primary background indicators (such as gender, age, geographic location, etc.) to generate demographic information association, that is, analyze the correlation between variables based on the basic demographic information of the subjects. For example, we analyze the differences in drug metabolism among subjects of different age groups and genders to reveal that demographic factors can affect clinical trial results.

[0047] Next, multi-source background information is subjected to multi-combination association analysis through secondary background indicators to generate more complex associations, such as clinical trial history associations and disease characteristics-drug response associations. By combining multiple background data, the relationship between the subject's previous trial records and current drug response, or the correlation between specific disease characteristics and drug response, is analyzed. For example, the blood sugar fluctuation pattern of diabetic patients in a specific drug trial can be identified to predict the response of similar patients.

[0048] In the completion phase, demographic information associations, clinical trial history associations, and disease feature-drug response associations are introduced into the completion generator to complete the missing background information. The completion generator is an algorithmic tool used to generate reasonable data for the missing parts based on existing data. Through the completion generator, these associations are used to complete the identified missing background information so that the data sets of all subjects are as non-missing as possible. For example, if a subject is missing drug metabolism gene information, the information can be inferred and completed based on the genetic data of other similar subjects.

[0049] The completed data needs to be verified. Based on the sample non-missing background information, sample missing background information and sample completed background information, a multi-source association analysis result is constructed to verify the accuracy of the completion. If the completed data passes the preset verification criteria, that is, meets the completion verification threshold, the completed background information is combined with the original non-missing background information to finally generate high-quality multi-source background information.

[0050] The multi-level background information processing and completion mechanism ensures accurate analysis of subject data from different sources. Even if some data are missing, reasonable data can be generated through completion, thereby improving the non-missing nature and reliability of clinical trial data.

[0051] Furthermore, the present application also includes:

[0052] The completed background information and the non-missing background information are verified for consistency across dimensions to generate a data logic verification result; based on the medical knowledge base and historical test data, the completed background information is verified for compliance with factual rules to generate a fact logic verification result; if both the data logic verification result and the fact logic verification result are qualified, the completed background information is obtained.

[0053] Specifically, two verification steps are used to ensure the accuracy of the completed background information. The first is cross-dimensional consistency verification, and the second is medical law compliance verification to ensure that the completed data is logically and medically reasonable and feasible.

[0054] Completion background information refers to the missing subject background data that has been generated through algorithmic completion to fill in the gaps in the original data. For example, if the drug metabolism gene information of a subject is missing, the system infers a reasonable complement information from the data of similar subjects.

[0055] Non-missing background information refers to the subject background data that is complete and does not need to be supplemented when the data is initially imported. For example, the subject's gender, age, basic health status and other data are complete and not missing when they are initially imported.

[0056] During the verification process, cross-dimensional consistency verification is first performed, that is, the completed background information is compared and analyzed with the non-missing background information to ensure that the completed data is logically consistent with the existing data. Cross-dimensional means that verification will be performed across multiple data dimensions (such as gender, age, medical history, etc.) to ensure that the relationship between dimensions is reasonable. For example, if a 60-year-old male subject shows that his medical history includes hypertension in the non-missing background information, then the completed drug metabolism gene information should be consistent with this medical history and should not produce drug metabolism characteristics that are inconsistent with hypertension. This verification step generates data logic verification results, indicating whether the completed data is logically consistent with the existing data.

[0057] Next, the medical knowledge base is a database containing a large amount of medical knowledge, disease characteristics, treatment methods, and drug reactions for reference and comparison. For example, it can contain the common effects of a certain drug on patients of different ages, or the typical course and symptoms of certain chronic diseases. Historical test data refers to the data accumulated from past clinical trials, including the actual reactions of the subjects, experimental results, and drug effects, etc., which provides the system with a real, long-term accumulated reference basis to help verify whether the completed data is reasonable. For example, a completed drug metabolism information can be compared with the historical data of similar patients to confirm its rationality. The completed background information is the subject background data inferred and completed based on the existing data, that is, the information that was originally missing in the data set. In order to ensure the accuracy of the completed data, it is verified by factual laws to check whether it conforms to known logic and laws. The factual logic verification result is the conclusion generated after verification, indicating whether the completed background information conforms to the existing medical laws and historical data. For example, the drug reaction data of a diabetic patient is completed, and the verification result may show that the completed data is consistent with the reactions of other similar patients, thereby confirming the rationality of the data. If the verification result shows that the completed data does not conform to the factual laws, the data may be further adjusted to ensure its accuracy.

[0058] Only when both the data logic verification result and the fact logic verification result are passed, that is, both results are verified to be qualified, will the supplementary background information be officially accepted. For example, if a subject's genetic data after completion shows that the subject has abnormal metabolism of a certain drug, and this abnormality is consistent with his medical history and factual laws, the supplementary information will be considered reliable, thus generating the final subject data set. Through these two verification steps, it is ensured that the supplemented data is not only logically reasonable, but also meets the known factual standards.

[0059] Furthermore, the present application also includes:

[0060] Generate multiple weak classifiers according to the primary background indicators and the secondary background indicators; adaptively adjust the weights of multiple weak classifiers according to the performance effect coefficient of the multi-source background information in the multi-source clinical trial data, obtain a strong classifier by weighting, and configure the strong classifier to the multiple initial cleaning rule generation strategies; extract multiple initial parameter combinations of the multiple initial cleaning rule generation strategies and perform fitness evaluation respectively, adjust the multiple initial parameter combinations based on the obtained multiple cleaning effect evaluation results, and obtain multiple standard parameter combinations; configure the multiple standard parameter combinations to obtain the multiple cleaning rule generation strategies.

[0061] Specifically, multiple weak classifiers are generated based on primary background indicators and secondary background indicators. Primary background indicators usually refer to basic demographic information, such as gender, age, geographic location, etc., and secondary background indicators cover more detailed health information, such as medical history, abnormal laboratory results, etc. Weak classifiers are some simple algorithm models that handle certain specific background data classification tasks alone. For example, one weak classifier may be specifically used to process the effect of age on drug response, and another may be used to process the effect of geographic location on health data.

[0062] Next, the weights of weak classifiers are automatically adjusted based on the performance of multi-source background information to generate a strong classifier. Multi-source background information refers to the performance of clinical trial data analysis of subjects that combines multi-dimensional background data of different subjects, such as age, health status, lifestyle, etc. The performance coefficient is used to measure the actual effect of this background information in clinical trial data analysis. For example, some background information has a greater impact on abnormal data detection. By adaptively adjusting the weights of weak classifiers, the weights of more useful classifiers are gradually increased, and the weights of poorly performing classifiers are weakened, thereby weightedly generating a strong classifier that can handle more complex problems. The strong classifier will eventually be applied to multiple initial cleaning rule generation strategies to help more effectively process data from different sources.

[0063] Next, multiple initial parameter combinations from multiple initial cleaning rule generation strategies are extracted, including the initial setting values ​​of various initial cleaning rule generation strategies, such as outlier thresholds, data consistency standards, etc. Fitness evaluation is used to measure the effects of these initial parameters and evaluate whether these parameters can effectively clean the data. Fitness evaluation refers to judging whether the performance of the current parameter combination is good enough through evaluation criteria, such as whether the cleaned data is consistent, whether key specific data is retained, etc. If the evaluation results are not ideal, adjust the initial parameter combination according to the evaluation results, and gradually optimize these parameters until the optimal standard parameter combination is found to maximize the cleaning effect.

[0064] Finally, the optimized standard parameter combination is configured into the cleaning process to generate multiple cleaning rule generation strategies. Multiple cleaning rule generation strategies refer to the generation strategies of cleaning rules that have been optimized and adapted, which can ensure the consistency and efficiency of data cleaning. For example, in the cleaning process, multiple cleaning rule generation strategies are applied to ensure that the steps of outlier processing and missing value filling of all data sources are carried out in accordance with the standards of each research center, ensuring that clinical trial data from different sources can be processed, optimizing the cleaning effect, and ensuring the reliability of data analysis.

[0065] Furthermore, the present application also includes:

[0066] Each initial parameter combination includes a different abnormality determination threshold generation strategy and a different specific data processing solution generation strategy.

[0067] Specifically, each initial parameter combination is a set of parameters set at the beginning of data processing and analysis to guide the identification of abnormal data and the processing of specific data. The initial parameter combination includes an abnormal determination threshold generation strategy and a specific data processing solution generation strategy.

[0068] First, the abnormal judgment threshold generation strategy refers to setting the threshold for judging abnormal data according to different rules or standards. The abnormal judgment threshold is a critical point. When the data exceeds the abnormal judgment threshold, it will be marked as abnormal. For example, in a clinical trial, when a subject's blood sugar level exceeds 180 mg per deciliter, it will be judged as abnormal data according to the abnormal judgment threshold. Different test conditions or subject groups may require different judgment strategies. For example, the blood pressure standards for the elderly and young people are different, so the system needs to generate different threshold strategies to adapt to different group characteristics.

[0069] Secondly, the strategy for generating specific data processing solutions refers to the processing solutions designed for data from subjects of specific groups or backgrounds. Specific data refers to data from subjects with special backgrounds or differences. For example, subjects from high altitude areas usually have higher red blood cell counts, or diabetic patients have larger blood sugar fluctuations. When processing, they cannot be simply cleaned according to general standards, but require special processing solutions. By generating different specific data processing strategies, it is ensured that key data is reasonably retained and processed during the analysis process and will not be misjudged as abnormal or lost.

[0070] For example, in a clinical trial, different red blood cell count criteria (i.e., different abnormality threshold generation strategies) may be set for subjects in low-altitude and high-altitude areas, and specific data processing plans may be designed for each subject group to ensure that the specific data of these groups are appropriately retained and processed.

[0071] Furthermore, the present application also includes:

[0072] A parameter fitness function is constructed through data consistency, anomaly rate and specific data retention rate; multiple parameter fitnesses of the multiple initial parameter combinations are calculated according to the parameter fitness function, and multiple initial parameter combinations corresponding to multiple parameter fitnesses exceeding multiple parameter fitness thresholds are extracted to obtain the multiple standard parameter combinations.

[0073] Specifically, in the process of optimizing data cleaning rules, a parameter fitness function is constructed through data consistency, anomaly rate, and specific data retention rate to evaluate and optimize the performance of the initial parameter combination. The parameter fitness function is a quantitative evaluation standard used to measure the effect and applicability of different parameter combinations in data cleaning.

[0074] Data consistency refers to the consistency and integration of cleaned data across sources and dimensions. For example, when processing clinical data from different research centers, it is necessary to ensure that all data are cleaned according to their respective standards so that the final data set can maintain a consistent structure and format.

[0075] The anomaly rate refers to the proportion of data identified as outliers during the data cleaning process. Calculate the proportion of abnormal data and reduce the misjudgment of outliers by adjusting the cleaning parameters. For example, if a cleaning rule is too strict, normal data may be misjudged as abnormal. You need to optimize the parameters to reduce this misjudgment.

[0076] The specific data retention rate refers to the proportion of key data with specific background that are retained during the cleaning process. It is necessary to ensure that data with special significance in clinical trials are not oversimplified or lost during the processing. For example, abnormal red blood cell counts in high altitude areas should be identified and retained, rather than being mistakenly deleted as common abnormal values.

[0077] The parameter fitness function constructed by the standard can calculate the parameter fitness of each initial parameter combination, reflecting the effect of the parameter combination in meeting the above standards. Then, the parameter combinations with parameter fitness higher than the preset parameter fitness threshold are extracted to find the best performing parameter combination, thereby ensuring that the final generated standardized cleaning rules are efficient and accurate.

[0078] Furthermore, the present application also includes:

[0079] Analyze similar abnormal subjects according to the multi-source background information to obtain similar abnormal subject data; extract the similar abnormal subject data from the multi-source clinical trial data for data comparison to obtain the local specific data; if the specific data amount of the local specific data is equal to 0, obtain the non-local specific data, and standardize the non-local specific data by the non-local cleaning rule generated by the cleaning rule generation strategy to obtain the non-local clinical trial data; if the specific data amount is greater than 0, obtain the local specific data, and standardize the local specific data by the local cleaning rule generated by the cleaning rule generation strategy to obtain the local clinical trial data; combine the non-local clinical trial data and the local clinical trial data to obtain clinical trial data.

[0080] Specifically, the subjects are analyzed based on multi-source background information to identify groups with similar abnormal characteristics, called similar abnormal subjects, and the data of these subjects, namely similar abnormal subject data, are extracted. Multi-source background information refers to data from different sources, such as the subject's gender, age, geographic location, medical history, etc. By analyzing this information, it is possible to find subjects with similar backgrounds and showing similar abnormal reactions, for example, patients from the same region who have similar adverse reactions to a certain drug.

[0081] Next, the data of these similar abnormal subjects are extracted from multi-source clinical trial data and compared to obtain local specific data. Multi-source clinical trial data includes laboratory test results, drug reactions, adverse event records, etc. When the comparison finds that the data of some subjects appear abnormal or have special characteristics in a specific context, these data are marked as local specific data. For example, the red blood cell count of some subjects in high altitude areas may be higher than that of subjects in other areas, which is local specific data.

[0082] If no specific data is found through comparison (i.e. the amount of specific data is equal to 0), it means that the data of all similar abnormal subjects behave consistently during the cleaning process, and non-local cleaning rules are generated. Non-local specific data refers to data that does not have significant individual differences, and these data can be processed by conventional cleaning standards. Non-local cleaning rules are rules generated based on conventional data processing methods, for example, standardized cleaning methods for health indicators such as blood pressure and blood sugar that are applicable to most subjects. Non-local specific data are standardized through these rules to obtain non-local clinical trial data.

[0083] However, if the amount of specific data is greater than 0 after comparison, these local specific data will be processed and local cleaning rules will be generated to adapt to data with special backgrounds or characteristics. For example, subjects in high-altitude areas have higher red blood cell counts, and specific cleaning rules are applied to standardize local specific data and generate local clinical trial data.

[0084] Ultimately, non-local clinical trial data and local clinical trial data are combined to generate complete clinical trial data, which ensures the uniformity of the data while retaining personalized treatment for specific groups, making the clinical trial data both widely applicable and able to reflect the unique circumstances of specific subjects.

[0085] Furthermore, the present application also includes:

[0086] The local specific data is judged to be consistent with the specific range; if the judgment result satisfies the specific range, the local specific data is retained; if the judgment result does not satisfy the specific range, the local specific data is locally optimized.

[0087] Specifically, by analyzing the local specific data, it is determined whether the local specific data is within the preset specific range, and then it is decided whether to retain or optimize the local specific data. First of all, local specific data refers to data that show significant differences in the subject population due to specific background factors (such as geographical location, medical history, etc.). For example, the red blood cell count of subjects in high altitude areas may be significantly higher than that of subjects in low altitude areas.

[0088] The specific range is a reference interval preset by the system to evaluate whether the local specific data is reasonable or needs further processing. The specific range is based on known medical knowledge or historical experience of clinical experimental data. For example, the system may preset the normal range of red blood cell counts in high altitude areas to be 5 million to 6 million per cubic millimeter. If the subject's red blood cell count is within this range, it is considered that the data meets the specific range.

[0089] When judging local specific data, if the results show that the data meets the requirements of the specific range, it means that the data is within the normal specific range, and the data is selected to be retained. In other words, the local specific data will be directly retained without modification or optimization as valid data for subsequent analysis. For example, the subject's red blood cell count is 5.5 million per cubic millimeter, which is judged to be normal under specific geographical conditions, so the data will be retained.

[0090] However, if the judgment result indicates that these data are not within the preset specificity range, they will be locally optimized. Local optimization means that the system fine-tunes or corrects these data that do not meet the specificity range to make them more in line with the actual situation. For example, if the red blood cell count of subjects in high altitude areas is 7 million per cubic millimeter, which is obviously beyond the preset specificity range, the value will be adjusted based on the data of other similar subjects to keep it within a reasonable range for subsequent analysis, ensuring that the abnormal values ​​of the data will not affect the analysis results while retaining the medically significant specificity.

[0091] By being able to accurately process specific data in different groups, unnecessary data modifications are avoided and the accuracy of the data during the standardization process is ensured.

[0092] In summary, the clinical trial data risk early warning system based on big data analysis provided by this application has the following technical effects:

[0093] A data import module is used to receive multi-source clinical trial data of multiple subjects imported by multiple research centers; a background information acquisition module is used to obtain multi-source background information of the multiple subjects; a dynamic adaptive cleaning module is used to adaptively optimize multiple initial cleaning rule generation strategies of the multiple research centers to obtain multiple cleaning rule generation strategies; a local specificity identification module is used to perform local specificity identification on the multi-source clinical trial data according to the multi-source background information to obtain local specific data, perform standardization processing on the non-local specific data through the non-local cleaning rules obtained based on the cleaning rule generation strategy to obtain non-local clinical trial data, perform standardization processing on the local specific data through the local cleaning rules obtained based on the cleaning rule generation strategy to obtain local clinical trial data, combine the non-local clinical trial data and the local clinical trial data, and perform standardization processing on the local specific data through the local cleaning rules obtained based on the cleaning rule generation strategy to obtain local clinical trial data. The clinical trial data is obtained by using clinical trial data from multiple sources, wherein the non-local clinical trial data is obtained by removing the local specific data from the multi-source clinical trial data; an early warning module is used to perform abnormality detection on the non-local clinical trial data and the local clinical trial data to obtain abnormal data, and to perform early warning analysis on the abnormal data to generate a real-time risk early warning report; a real-time risk adjustment module is used to identify dynamic changes in the real-time risk early warning report, and to adjust the early warning level in real time. In other words, the clinical trial data risk early warning system based on big data analysis is a highly integrated clinical trial data processing platform, which includes multiple modules that work together to ensure data accuracy and real-time risk management, and achieves the technical goal of rapid processing of data across multiple research centers, thereby achieving the technical effect of improving data integration accuracy, ensuring accurate analysis results and effectively retaining local specific data.

[0094] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0095] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. A clinical trial data risk early warning system based on big data analysis, characterized by: include: A data import module, used to receive multi-source clinical trial data of multiple subjects imported by multiple research centers; A background information acquisition module, used to acquire multi-source background information of the multiple subjects, where the background information refers to various basic data related to the individual subjects; A dynamic adaptive cleaning module, used for adaptively optimizing a plurality of initial cleaning rule generation strategies of the plurality of research centers to obtain a plurality of cleaning rule generation strategies; A local specificity identification module, used for performing local specificity identification on the multi-source clinical trial data according to the multi-source background information to obtain local specificity data, performing standardization processing on the non-local specific data by the non-local cleaning rule obtained based on the cleaning rule generation strategy to obtain non-local clinical trial data, performing standardization processing on the local specific data by the local cleaning rule obtained based on the cleaning rule generation strategy to obtain local clinical trial data, and combining the non-local clinical trial data and the local clinical trial data to obtain clinical trial data, wherein the non-local clinical trial data is obtained by removing the local specific data from the multi-source clinical trial data, and refers to data without significant individual differences; the local specific data refers to data found by comparison that the data of the subject appears abnormal or has special characteristics in a specific background; An early warning module, used for performing anomaly detection on the non-local clinical trial data and the local clinical trial data to obtain abnormal data, and performing early warning analysis on the abnormal data to generate a real-time risk early warning report; A real-time risk adjustment module, used to identify dynamic changes in the real-time risk warning report and adjust the warning level in real time; Wherein, obtaining multi-source background information of the multiple subjects includes: Receiving gender information, age information and geographic location information of the multiple subjects to generate primary background information; Receiving basic health status information, chronic disease history information, abnormal laboratory test result information, and drug metabolism gene information of the multiple subjects to generate secondary background information; Combining the primary background information and the secondary background information to obtain the multi-source background information; The step of obtaining multiple cleaning rule generation strategies includes: Generate multiple weak classifiers according to the primary background indicators and the secondary background indicators; Adaptively adjust the weights of multiple weak classifiers according to the performance effect coefficient of the multi-source background information in the multi-source clinical trial data, obtain a strong classifier by weight, and configure the strong classifier to the multiple initial cleaning rule generation strategies; Extracting multiple initial parameter combinations of the multiple initial cleaning rule generation strategies and performing fitness evaluation respectively, adjusting the multiple initial parameter combinations based on the obtained multiple cleaning effect evaluation results, and obtaining multiple standard parameter combinations; The plurality of standard parameter combinations are configured to obtain the plurality of cleaning rule generation strategies.

2. The clinical trial data risk early warning system based on big data analysis according to claim 1, characterized in that: The acquiring of multi-source background information of the plurality of subjects further comprises: Identify missing background information from the multi-source background information based on the primary background indicators and the secondary background indicators to obtain missing background information, and screen the multi-source background information based on the missing background information to obtain non-missing background information, wherein the primary background information is the basic attributes of the subject, and the secondary background information reflects the long-term health status of the subject and its potential response to the drug; Performing correlation analysis on the multi-source background information through the primary background indicators to generate demographic information correlation; Performing multi-combination association analysis on multi-source background information through the secondary background indicators to generate clinical trial history associations and disease feature-drug response associations; Introducing the demographic information association, the clinical trial history association and the disease feature-drug response association into a completion generator, and performing missing completion on the non-missing background information and the missing background information through the completion generator to obtain the completed background information; A multi-source association analysis result is constructed according to the sample non-missing background information, the sample missing background information and the sample completed background information. The completed background information is verified based on the multi-source association analysis result. If the completion verification threshold is met, the multi-source background information is obtained by combining the completed background information and the non-missing background information.

3. The clinical trial data risk early warning system based on big data analysis according to claim 2, characterized in that: Before obtaining the multi-source background information by combining the supplemented background information and the non-missing background information, the method includes: Performing cross-dimensional consistency verification on the completed background information and the non-missing background information to generate a data logic verification result; Based on the medical knowledge base and historical test data, the supplemented background information is verified to be consistent with the fact rules, and a fact logic verification result is generated; If both the data logic verification result and the fact logic verification result are qualified, the supplementary background information is obtained.

4. The clinical trial data risk early warning system based on big data analysis according to claim 1, characterized in that: Each initial parameter combination includes a different abnormality determination threshold generation strategy and a different specific data processing solution generation strategy.

5. The clinical trial data risk early warning system based on big data analysis according to claim 1, characterized in that: The obtaining of multiple standard parameter combinations includes: The parameter fitness function is constructed by data consistency, anomaly rate and specific data retention rate; The multiple parameter fitnesses of the multiple initial parameter combinations are calculated according to the parameter fitness function, and the multiple initial parameter combinations corresponding to the multiple parameter fitnesses exceeding multiple parameter fitness thresholds are extracted to obtain the multiple standard parameter combinations.

6. The clinical trial data risk early warning system based on big data analysis according to claim 1, characterized in that: The clinical trial data obtained includes: Analyze similar abnormal subjects according to the multi-source background information to obtain similar abnormal subjects data; Extracting the abnormal subject data of the same type from the multi-source clinical trial data for data comparison to obtain the local specific data; If the specific data amount of the local specific data is equal to 0, the non-local specific data is obtained, and the non-local specific data is standardized by the non-local cleaning rule generated by the cleaning rule generation strategy to obtain the non-local clinical trial data; If the amount of specific data is greater than 0, the local specific data is obtained, and the local specific data is standardized by the local cleaning rule generated by the cleaning rule generation strategy to obtain the local clinical trial data; The non-local clinical trial data and the local clinical trial data are combined to obtain clinical trial data.

7. The clinical trial data risk early warning system based on big data analysis according to claim 6, characterized in that: Before obtaining the local clinical trial data, including: Performing a conformity judgment on the specific range of the local specific data; If the judgment result meets the specificity range, the local specificity data is retained; If the judgment result does not satisfy the specificity range, the local specificity data is locally optimized.

Images