Clinical test data verification method and system based on modular architecture

Abstract

The invention provides a clinical test data checking method and system based on a modular architecture, and the method comprises the steps: obtaining a target clinical test scheme document, and mapping the target clinical test scheme document into a term mapping table; on the basis of the protocol structure atlas, combined with test actual acquisition data, generating an execution acquisition atlas and path confidence; comparing the protocol atlas with the execution acquisition atlas, and identifying the structure deviation of the acquisition path through structure alignment check to generate a structure deviation matrix; obtaining an index basic fluctuation interval and a medical standard, and generating a consistency behavior score in combination with the generated structure deviation matrix to quantify a behavior trend anomaly score of the subject; and in combination with the protocol structure atlas, executing an acquisition atlas, a structure deviation matrix and a consistency behavior score, calculating a causal deviation score of each subject, and marking missing acquisition, wrong acquisition and overtime acquisition events in clinical test data according to the causal deviation scores.

Description

Technical Field

[0001] The present invention belongs to the field of clinical trial data verification, and particularly relates to a clinical trial data verification method and system based on a modular architecture. Background Art

[0002] As an indispensable link in new drug research and development, device evaluation, and treatment strategy verification, the data quality of clinical trials directly determines the scientificity, compliance, and regulatory pass rate of the entire trial. With the gradual mainstreaming of multi-center, cross-regional, and large-scale clinical trials, the trial data presents characteristics such as complex structure, diverse processes, long collection chains, and scattered responsible entities, making the data verification work in clinical trials particularly critical and complex. Currently, clinical trial data verification mainly relies on three traditional methods: one is manual CRF verification, that is, through CRA offline or remotely accessing the collected data and original documents for manual verification; the second is an electronic data verification system based on a rule engine, which automatically marks abnormal items by setting field-level verification rules; the third is that some research units attempt to use data visualization means to discover deviations or trend anomalies from a statistical perspective. These means have played an important role in the historical process of data verification, but they also face increasingly prominent limitations.

[0003] First of all, with the increasing complexity of clinical trial protocols, traditional data verification methods often can only focus on the field values themselves and cannot identify the "structural deviations" from the trial design. For example, whether a certain collection is executed on the correct population at the correct time according to the protocol preset process, and it is difficult for existing systems to systematically restore and verify the consistency between the protocol process and data execution. Secondly, in terms of data authenticity, current anomaly detection methods usually perform single-point anomaly judgment based on set thresholds or statistical models (such as mean variance, z-score, etc.), ignoring the temporality and internal behavior patterns of individual continuous records, resulting in some more concealed data forgery, supplementary recording, and misrecording behaviors that cannot be identified. In addition, clinical trials are essentially constructed around the causal mechanism of "event - response", such as the physiological responses that should occur after drug administration, the recovery indicators that should be observed after surgery, etc. These causal logics have not been modeled by the current data verification system, resulting in the situation that even if the data is normal in terms of numerical values, it may be completely unreasonable in terms of medical logic and is difficult to be detected by the system. Furthermore, the existing verification mechanism lacks a flexible modular architecture, resulting in a large amount of manual adjustment and configuration required when the trial protocol changes, multi-center deployment, or new indicators are introduced, and does not have the system capabilities of being combinable and adaptable. Therefore, how to construct an efficient data verification system that can not only understand the trial protocol process, but also combine data behavior patterns and causal logics, and at the same time has a modular structure has become an important technical issue that urgently needs to be solved in current clinical trial data management. Summary of the Invention

[0004] The object of the present invention is to design a clinical trial data verification method and system based on a modular architecture, propose a clinical trial data verification method and system based on a modular architecture, and establish an extensible and combinable data verification mechanism, which can realize intelligent, structured, and multi-dimensional verification of clinical trial data at different levels.

[0005] To achieve the above object, in the first aspect of the present invention, there is provided a clinical trial data verification method based on a modular architecture, and the method includes:

[0006] Obtain a target clinical trial protocol document, generate a protocol structure graph according to the time points, collection indicators, and subject populations of the target clinical trial protocol document, and map it into a term mapping table;

[0007] Based on the protocol structure graph, combine the actually collected data of the trial to generate an execution collection graph and path confidence;

[0008] Compare the protocol graph with the execution collection graph, identify the structural deviations of the collection path through structural alignment verification, so as to generate a structural deviation matrix;

[0009] Obtain the basic fluctuation range of indicators and medical standards, combine the generated structural deviation matrix, and generate a consistency behavior score for quantifying the abnormal score of the behavior trend of the subjects;

[0010] Combine the protocol structure graph, execution collection graph, structural deviation matrix, and consistency behavior score, calculate the causal deviation score of each subject, and mark the missed collection, miscollection, and overtime collection events in the clinical trial data according to the causal deviation score.

[0011] Further, the obtaining of the target clinical trial protocol document, generating a protocol structure graph according to the time points, collection indicators, and subject populations of the target clinical trial protocol document, and mapping it into a term mapping table specifically includes:

[0012] Obtain a target clinical trial protocol document; input the target clinical trial protocol document into a BERT-based model, perform structural label prediction on each paragraph, and output a structural quadruple, including time points, collection indicators, subject populations, and additional conditions in the original protocol that restrict the current collection behavior;

[0013] Convert the time points, collection indicators, and subject populations into three nodes and two edges to form a directed path from time point → subject population → collection indicator;

[0014] Map all nodes to a standard coding system as a term mapping table; wherein the collection indicators use LOINC; the subject populations use the ICF grouping structure; the time points are converted into the structure of visit days and allowable offset ranges.

[0015] Furthermore, the protocol structure graph also has an edge weight regularization mechanism, specifically including:

[0016] Extract the additional conditions restricting the current collection behavior from the original protocol in the target clinical trial protocol document, send the additional conditions restricting the current collection behavior in the original protocol into the BERT model to obtain the attention weight distribution of each token, calculate its entropy value as the semantic concentration index; and constrain the edge weight according to the semantic concentration index.

[0017] Furthermore, each piece of data in the actually collected data of the trial includes the actual collection time, the actual index name, and the actual subject ID;

[0018] Then,

[0019] Based on the protocol structure graph, combined with the actually collected data of the trial, generate an execution collection graph and path confidence, specifically including:

[0020] Perform term standardization processing on the actual index name based on the term mapping table to generate actual standardized terms; perform time alignment processing on the actual collection time to obtain the aligned actual time;

[0021] Construct an execution collection graph based on the results of term standardization and time alignment processing, as well as the actual subject ID, and calculate the path confidence; wherein, the execution collection graph includes several directed collection paths, the directed path of the aligned actual time → actual subject ID → actual standardized term; the path confidence is jointly determined by the time matching degree and semantic consistency; the time matching degree is determined based on the actual time node and time point; the semantic consistency is determined by the standardized index node and the collection index.

[0022] Furthermore, the execution collection graph also has a local protocol coverage penalty term for controlling the deviation between the number of actually collected paths and the number specified in the protocol at a certain time point.

[0023] Furthermore, compare the protocol graph and the execution collection graph, and identify the structural deviation of the collection path through structural alignment verification to generate a structural deviation matrix, specifically including:

[0024] Obtain the paths of the protocol graph and the execution collection graph, and perform structural similarity analysis; wherein the structural similarity analysis includes node semantic consistency, population consistency, and time tolerance matching; if the protocol graph meets the structural similarity analysis, select the one with the smallest deviation degree as the optimal matching path;

[0025] Output the minimum deviation degree of all paths of the protocol graph and form a structural deviation matrix;

[0026] Wherein, the semantic consistency of the node means that there is a mapping between the collected index and the actual standardized term in the term mapping table, or the cosine similarity ≥ 0.8; the population consistency means that the actual subject ID is the tested population; the time tolerance matching means that the actual collection time difference does not exceed ±2 days.

[0027] Further, the deviation degree is calculated as follows:

[0028] ;

[0029] Where, is the deviation degree, is the confidence level of the execution collection map path; is the collection index of the protocol map of the local coverage penalty term, indicating the over-collection / under-collection degree of the collection structure; is the offset aggregation regularization term, measuring whether there are a large number of non-standard paths around this time point; , are the regularization term adjustment coefficients.

[0030] Further, the obtained index basic fluctuation interval and the medical standard are combined with the generated structure deviation matrix to generate a consistency behavior score for quantifying the abnormal score of the subject's behavior trend, specifically including:

[0031] Extract the time series data set of each subject from all the collection paths of the execution collection map path;

[0032] Construct the behavior trend tensor of the subject according to the time series data set for depicting the change rate pattern at adjacent time points;

[0033] Calculate the trend perturbation density regularization term based on the behavior trend tensor and the time series data set , expressed as:

[0034] ;

[0035] Where, T is the total time window, t is the current time window, represents the variance of the change rates of all indicators at the th time point, is the correlation between two indicator time series; is the expected correlation of this pair of indicators in the normal tested population; is the indicator collaborative deviation adjustment factor;

[0036] Combine the trend perturbation density regularization term and the protocol structure deviation matrix to obtain the final residual term; determine the consistency behavior score through the final residual term.

[0037] Further, calculating the causal deviation score of each subject , which is expressed as:

[0038] ;

[0039] where represents whether the subject meets the th causal trigger condition; is the total acquisition window, is the behavior anomaly threshold; represents the acquisition confidence of the causal path in ; is the structure-behavior joint weight;

[0040] where represents that the condition holds but the acquisition is weak or missing; represents that the condition does not hold but the acquisition is still carried out and the behavior is abnormal.

[0041] In the second aspect of the present invention, a clinical trial data verification system based on a modular architecture is provided. The system includes:

[0042] A protocol parsing module for obtaining a target clinical trial protocol document, generating a protocol structure graph according to the time points, acquisition indicators, and subject populations of the target clinical trial protocol document, and mapping it into a term mapping table;

[0043] A data fusion module for generating an execution acquisition graph and path confidence based on the protocol structure graph and combining the actual acquisition data of the trial;

[0044] A structure verification module for comparing the protocol graph with the execution acquisition graph, identifying the structural deviation of the acquisition path through structural alignment verification, and generating a structure deviation matrix;

[0045] A behavior modeling module for obtaining the basic fluctuation range of indicators and medical standards, combining the generated structure deviation matrix, and generating a consistency behavior score for quantifying the abnormal score of the behavior trend of the subject;

[0046] A causal reasoning module for combining the protocol structure graph, the execution acquisition graph, the structure deviation matrix, and the consistency behavior score, calculating the causal deviation score of each subject, and marking the missed acquisition, mis-acquisition, and timeout acquisition events in the clinical trial data according to the causal deviation score.

[0047] The beneficial technical effects of the present invention are at least as follows:

[0048] In view of the problems of weak structural process verification ability, rough behavior pattern analysis, lack of causal logic, and poor system scalability in the existing technology, the present invention proposes a clinical trial data verification method and system based on a modular architecture, and establishes an extensible and combinable data verification mechanism, which can realize intelligent, structured, and multi-dimensional verification of clinical trial data at different levels. In the system design of the present invention, the modular construction idea is adopted, and the data verification process is disassembled into several sub-modules with independent functions and standardized interfaces, which can be flexibly combined and dynamically adapted to changes according to the actual needs of the test plan; in the verification logic, it is proposed to carry out collaborative analysis from three dimensions of "protocol process execution consistency", "intra-individual behavior pattern consistency", and "medical causal logic consistency", so as to construct a multi-level closed-loop verification system from planning to execution, from structure to content, and from rules to reasoning. Through this design, the present invention not only realizes the accurate identification of structural deviations, potential forgeries, and medical anomalies, but also enhances the system's adaptability to protocol version adjustment, multi-center deployment, and large-sample parallel processing scenarios, significantly improving the efficiency and automation level of clinical trial data quality control, and providing a solid data infrastructure support for future digital and intelligent clinical trials. Description of the Drawings

[0049] The present invention is further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation to the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the following drawings.

[0050] Figure 1 It is a flowchart of the clinical trial data verification method based on the modular architecture of the present invention. Detailed Embodiments

[0051] The embodiments of the present invention are described in detail below. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements with the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary only for explaining the present invention and should not be construed as limiting the present invention.

[0052] In one or more embodiments, as Figure 1 shown, a clinical trial data verification method based on a modular architecture is disclosed, and the method includes the following S1 to S5:

[0053] S1. Obtain the target clinical trial protocol document, generate a protocol structure map according to the time points, collection indicators, and subject populations of the target clinical trial protocol document, and map it into a term mapping table.

[0054] Specifically, this step aims to extract and structure the content regarding collection time points, indicator items, and the subject population in the clinical trial protocol document, and construct a protocol graph for unified use in subsequent processes. .

[0055] Since clinical trial protocols are often written in natural language, they have problems such as inconsistent language styles, complex nested collection logics, frequent conditional collections, and many synonymous terms, and cannot be directly processed by keyword extraction or rule template methods. Therefore, the present invention designs a multi-task structure recognition model to classify the structural intentions and identify term activations of protocol paragraphs, and at the same time combines semantic fuzzy constraints when generating the graph to generate a protocol graph with medical logic consistency and semantic expression completeness.

[0056] The input of the model is the segmented text of the protocol document , and each represents a semantic segment, usually at the natural paragraph level. The present invention first uses a BERT-based model to predict the structural labels of each paragraph and at the same time identify whether it triggers the keywords in the standard term library. This model has two output heads: one is a classifier for paragraph types (output structural categories: time points, indicator items, population conditions, or irrelevant content), and the other is a binary classifier for term activation. The loss function for joint training is as follows:

[0057] ;

[0058] where represents the classification cross-entropy loss of the structural type, represents the binary cross-entropy loss of the term activation annotation, is the balance parameter, set to 0.5. This model uses BERT-base as the backbone, with each layer output being 768-dimensional, and two sets of fully connected layers are used for classification after the output. The term library in the input is used as the supervision signal for trigger word detection.

[0059] The structural labels and term hit results output by the model will be aggregated into structural quadruples to form the following set:

[0060] ;

[0061] Each quadruple represents the definition of a collection event. Among them is the time point description (such as "Day14"), is the collection indicator (such as "ALT"), is the subject population (such as "intervention group"), It is an additional condition in the original agreement to restrict this collection behavior. To enhance operability, the present invention introduces a sliding window mechanism in the processing to establish the relevance between paragraphs. For example, if the th paragraph states "Day7", and the th paragraph mentions "collect ALT and WBC", the model will automatically aggregate these two paragraphs into a quadruple .

[0062] Among them, the fourth element does not participate in the construction of the graph nodes, but only serves as a conditional attribute of the graph edge, which is used for fuzzy edge weight, collection dependency, and causal relationship extraction.

[0063] Take the following protocol fragment as an example:

[0064] "On Day7, it is necessary to collect indicators such as ALT and WBC for all subjects in the intervention group. If the patient has an adverse reaction, an additional electrocardiogram (ECG) is also required."

[0065] The model will identify two quadruples:

[0066] ;

[0067] ;

[0068] Then it enters the graph generation stage. The present invention uses the following function to convert the set of quadruples into a protocol graph (the fourth element does not participate in the construction of the graph nodes):

[0069] ;

[0070] Each quadruple in the graph is converted into three nodes (time point, subject population, collection indicator) and two edges, forming a directed path from → → To express the uncertainty of conditional logic, the present invention designs an edge weight regularization mechanism: the conditional sentence extracted from the original text is sent to the BERT model to obtain the attention weight distribution of each token , and its entropy value is calculated as a semantic concentration index. The edge weight is defined as follows:

[0071] ;

[0072] Among them, represents the attention distribution entropy, and is a regulation parameter (set to 0.6). For example, when the conditional sentence is "if there is an adverse reaction, an additional ECG can be added", the attention is concentrated on fuzzy words such as "can be added", and the entropy value is relatively high, If it is lower, this edge will be marked as a "low - certainty edge", and its alignment requirements will be appropriately weakened during subsequent graph matching. This semantic mapping mechanism based on text entropy is an innovative structured expression of the semantic ambiguity problems existing in clinical protocols.

[0073] Finally, all nodes will be mapped to a standard coding system, (for indicators) using LOINC, (for populations) using the ICF grouping structure, and will be uniformly transformed into the structure of visit date + allowed offset range. The graph nodes and edges will record the original text index, protocol version number, and sentence element position to support subsequent cross - version comparison.

[0074] S2. Based on the protocol - structured graph described above, combined with the actual data collected in the trial, generate an execution - collection graph and path confidence.

[0075] Specifically, the goal of this step is to start from the actual data collected and, combined with the protocol - structured graph generated in the previous step and the standard terminology mapping table , construct an execution - collection graph that is consistent with the protocol structure and has the ability to express real - world data . This graph is the key bridge in the entire patent system to transform "trial design" into "data execution". It is not only used for the structural consistency analysis in step three but also serves as the behavioral information source for subsequent behavior modeling and causal logic analysis. Its quality will directly determine the accuracy and robustness of the verification process.

[0076] Compared with the protocol structure, the actual data collected in clinical trials has a high degree of uncertainty, including:

[0077] Time - point offset (e.g., the sample collection on Day 7 occurs on Day 6 or Day 9);

[0078] Inconsistent indicator names ("alanine aminotransferase" vs "ALT");

[0079] Subject numbers deviate from the population grouping;

[0080] Multiple collections of the same indicator (e.g., due to data loss and subsequent re - collection);

[0081] Central differences lead to inconsistent data structures and field encodings.

[0082] Therefore, the present invention proposes a two - layer mapping mechanism based on graph - structure reconstruction + collection - credibility regularization, and at the same time introduces a local protocol - coverage penalty term to control the problem of graph - structure imbalance caused by "over - collection", thereby constructing a graph with dynamic fault - tolerance ability, structure - alignment ability, and behavior - traceability ability .

[0083] The inputs for this step include:

[0084] Protocol structure map , from S1; Term mapping table , generated by S1 for indicator standardization; Actual acquisition record set , each , representing the actual acquisition time, actual indicator name, and actual subject ID.

[0085] Furthermore, first for all Utilize to perform term standardization. If is not in the standard term set, then use the BERT+CRF joint named entity recognition model to extract keywords, and then use cosine similarity to complete fuzzy matching, and record the matching distance as the semantic consistency weight , with a value range , the closer to 1 indicates that the indicator name is consistent with the protocol, and the actual standardized term is obtained .

[0086] Then for all Use the protocol map The time nodes defined in to perform time alignment processing to generate the aligned actual time. The alignment method is based on an asymmetric time distance function to avoid early acquisitions being misidentified as normal:

[0087] ;

[0088] Among them, is the actual acquisition time, is the standard date corresponding to the protocol design time point, is the adjustment factor (default 0.4). This design emphasizes the realistic clinical preference of "delayed acquisitions are tolerable, early acquisitions require caution", reflecting the semantic embedded understanding of actual acquisition behavior.

[0089] After completing term and time alignment, the present invention constructs the nodes and path edges of the execution acquisition map . Each acquisition record corresponds to an acquisition path in the graph, which consists of the time node , the subject node , and the standardized indicator node to form a directed path . The path confidence is jointly determined by the time matching degree and semantic consistency:

[0090] ;

[0091] This confidence is recorded as an edge attribute on the edge For subsequent atlas comparison, "soft-edge matching" is performed.

[0092] Furthermore, the present invention designs a penalty term for the structural distortion problem caused by "repeated collection" or "out-of-scope collection" in clinical trials, namely the local protocol coverage penalty term , which is used to control the deviation between the actual number of collection paths at a certain time point

[0093] ;

[0094] Among them, is the number of actually connected to the node (the type of indicators actually collected at a certain time node in the execution diagram), is the number of corresponding indicators to be collected in the protocol diagram. This item is used as a structural regularization signal in the subsequent behavior scoring model to avoid the situation of "central habit of over-collection" without marking protocol changes.

[0095] Taking specific data as an example:

[0096] If ALT and WBC are to be collected on Day 7 in the protocol, and in the record uploaded by a certain center, patient P001 collected ALT on Day 9 and AST, WBC, CRP, GLU on Day 7, the system processing process is as follows:

[0097] ALT and Day 9 → Delayed collection, , but the terms are consistent ;

[0098] WBC and Day 7 → Normal collection, ;

[0099] AST, CRP, GLU have no protocol mapping, low confidence or classified into , and semantic matching controls whether to retain.

[0100] Finally, when constructing , high-confidence paths (such as ALT, WBC) will be retained, fuzzy paths (such as AST) will be marked, and will be generated at the Day 7 node,

[0101] indicating that there may be "over-collection". The final output includes: the execution collection atlas , all collection paths and their edge weights; the set of local protocol coverage penalty terms

[0102] S3. Compare the protocol graph with the execution collection graph, and identify the structural deviation of the collection path through structural alignment verification to generate a structural deviation matrix.

[0103] Specifically, the goal of this step is to perform structural alignment verification on the protocol structure graph output in Step 1 and the execution collection graph output in Step 2 to identify whether the actual execution is consistent with the protocol plan, whether there are behavioral deviations such as "not collected", "wrongly collected", "early or delayed collection", "non-structural redundant collection", etc., and finally construct a structural deviation matrix

[0104] as the output.

[0105] It can be understood that the structural consistency verification plays a key hub role in the entire patent solution and is the core comparison bridge between "static design" and "dynamic execution". Its accuracy not only affects the input basis of the subsequent behavior model and causal path, but also directly affects the operability of the entire verification system. Three innovative mechanisms are introduced in the design of this step: (1) Structural path fuzzy matching mechanism; (2) Joint modeling of confidence and edge coverage penalty; (3) Collection path aggregation drift regularization term, which is used to solve various structural deviations caused by the fuzzy expression of protocol language, data upload delay and different collection habits between centers in clinical trials.

[0105] Furthermore, the present invention first establishes an alignment matrix at the path level. For each path in the protocol graph , the system searches for a structurally similar path in the execution graph.

[0106] The criteria for structural similarity consist of the following three conditions: and have a mapping in the term mapping table, or the cosine similarity ;

[0107] Population consistency: , that is, the same population;

[0108] Time tolerance matching: , indicating that the collection time difference does not exceed ±2 days.

[0109] If multiple paths meet the above criteria, the system selects the one with the minimum deviation function as the optimal matching path, and the deviation function is as follows:

[0110] ;

[0111] Where: is the degree of deviation,

[0112] is Confidence level (given by Step 2, representing the execution quality);

[0113] is a protocol node The local coverage penalty term, indicating the "over - sampling / under - sampling degree" of the acquisition structure;

[0114] is the offset aggregation regularization term, measuring whether there are a large number of non - standard paths around this time point (for example, a large number of "nearby" metrics are collected intensively);

[0115] , are the regularization coefficient adjustment factors (recommended to be set to 0.3 and 0.2).

[0116] Offset aggregation regularization term The calculation method is as follows:

[0117] ;

[0118] Where:

[0119] represents the protocol time point All acquisition paths within ±2 days around it;

[0120] is the confidence level of each path;

[0121] is the threshold (such as 0.6), indicating "low - confidence paths";

[0122] This item counts the "density of suspicious acquisition paths". If there are abnormally many aggregations, it indicates that there may be systematic offsets or non - structural expansion behaviors at this node.

[0123] Finally, the system outputs the minimum deviation degree for each path in the protocol graph to form a structural deviation matrix:

[0124] ;

[0125] represents the protocol path The degree of deviation from its minimum - structure - difference path among all execution paths. The higher the value, the more serious the deviation. The system inputs this matrix as a structural - risk heat map into the downstream model.

[0126] Taking a specific scenario as an example: If ALT and WBC are planned to be collected on Day 14 in but ALT appears on Day 16 in ), WBC appears on Day 13 and multiple unplanned indicators such as GLU, AST, and CRP are also collected (resulting in ), and there are 8 paths with a confidence level < 0.5 within this time window ( ), then finally for this path , the system automatically determines it as a "high structural deviation path".

[0127] The output includes: a structural deviation matrix : Each protocol path corresponds to a deviation value.

[0128] S4. Obtain the basic fluctuation range of indicators and medical standards, and combine with the generated structural deviation matrix to generate a consistency behavior score for quantifying the abnormal score of the behavior trend of the subject.

[0129] Specifically, the goal of this step is to construct a behavior consistency scoring mechanism that integrates structural deviation-driven, trend trajectory analysis, and indicator collaborative anomaly modeling , for quantifying the degree of behavior deviation of each subject during the actual collection process. Behavior deviation is not equivalent to the numerical abnormality of a certain indicator, but refers to whether the indicator fluctuations, trend directions, and the linkage between indicators are "reasonable", "stable", and "in line with the medical change logic under the test background" within a period of time. Different from traditional anomaly detection methods such as z-score and sliding window outlier detection, this solution completely conducts behavior-level modeling based on the structural graph, execution graph, and structural scoring results in the previous steps, reflecting the coupled modeling idea under the systematic patent structure.

[0130] Among them, the input of this step includes: the execution collection graph : from step 2, including the collection behavior, actual indicator values, and path confidence levels of each subject ; the protocol structural deviation matrix : from step 3, used to provide the focus of structural anomalies; the standard term mapping table : used to normalize the indicator values of each center; the basic fluctuation range of indicators and medical standards (provided by the clinical knowledge base), used for normalization and trend anomaly modeling.

[0131] Furthermore, the present invention first extracts from all collection paths the time series data set of each subject , where represents the th indicator collected by this subject at time point . All indicator values are matched via , mapped to the standard code and unified in units, and normalized to the [0, 1] interval.​

[0132] Then construct the subject behavior trend tensor , which is used to characterize the change rate pattern at adjacent time points:

[0133] ;

[0134] Where:

[0135] represents the value of the th index at time ;

[0136] is the standard fluctuation value of this index in the subject population (extracted from the clinical knowledge base);

[0137] is a small constant to avoid division by zero (such as );

[0138] This tensor is expanded into a structure in the time dimension and the index dimension .

[0139] Considering that some abnormalities in real clinical data are not caused by mutations, but are caused by the collective "shift" or "abnormal stability" of indicators (such as maintaining at the marginal value for a long time, loss of cooperation between indicators), the present invention introduces a trend perturbation density regularization term , which is used to capture the behavioral state of "obviously non - concentrated change rate":

[0140] ;

[0141] Where:

[0142] represents the variance of the change rates of all indicators at the th time point. A low value represents that the fluctuations of all indicators converge, and too low may be "abnormal stability";

[0143] is the correlation between two indicator time series;

[0144] is the expected correlation of this pair of indicators in the normal subject population;

[0145] is the indicator cooperation deviation adjustment factor, with an empirical value of 0.3;

[0146] This regularization term can both punish the "false stability" with too concentrated fluctuations and punish the structural abnormalities under "indicator cooperation destruction".

[0147] Final residual term Integrate the degree of trend change and collaborative perturbation, and introduce the structural deviation weight As the attention area adjustment signal, construct the consistency behavior score:

[0148] ;

[0149] Wherein,

[0150] The first term represents the overall fluctuation intensity of the subject in all time periods;

[0151] The second term activates the trend perturbation regularization in the structurally abnormal area to further punish atypical trend behaviors;

[0152] is the regularization adjustment coefficient (empirical value 0.5 - 1.0);

[0153] is the structural deviation activation threshold (such as 0.7).

[0154] The system finally outputs the behavior consistency score , normalized to (0, 1), the higher the value, the more "in line with the medical trend logic", and the lower the value, the stronger the behavior deviation. This value will be directly used in the subsequent causal path activation and risk fusion scoring.

[0155] For example: During the collection in a certain center, for a subject, the changes in three indicators of ALT, WBC, and CRP from Day 5 to Day 12 were extremely small (<2%), the fluctuation variance <0.005, and these time points were all at a high level of structural deviation ( ), the system immediately judged it as "false stability under structural abnormality", and through activating the trend perturbation penalty, resulting in significantly decreasing (for example ), and thus being classified as an individual with a high risk of abnormal behavior.

[0156] The output of this step is: : The behavior consistency score of each subject; Optional: Mark the behavior abnormality degree map according to the time point and indicator dimension for the visualization system to display.

[0157] S5. Combine the protocol structure map, execution collection map, structural deviation matrix, and consistency behavior score, calculate the causal deviation score of each subject, and mark the missed collection, mis - collection, and overtime collection events in the clinical trial data according to the causal deviation score.

[0158] Specifically, the goal of this step is to complete the identification and modeling of the causal logic deviation existing in the clinical trial execution process, and output the causal consistency deviation score of each subject , and clearly mark the breakpoint of the causal path as the direct basis for subsequent automatic system review, manual review by auditors, and report generation.

[0159] In the previous steps, the system has completed structure alignment (step 3) and behavior consistency modeling (step 4). However, the compliance of clinical trials not only requires "correct structure acquisition" and "data within limits", but more importantly, whether logical acquisition has been carried out under the correct preconditions, that is, causal consistency.

[0160] For example, the protocol defines: "If a serious adverse event (SAE) occurs, additional blood pressure and ECG indicators need to be collected within 48 hours." Such acquisition designs with pre-logical conditions are extremely common, but in actual execution, the following may occur:

[0161] The triggering condition is met but not collected (omission);

[0162] The triggering condition is not met but still collected (wrong collection);

[0163] Collected when the condition is met but overtime, wrong indicators collected, abnormal behavior;

[0164] This step is exactly used to identify and quantify these causal logic misalignment behaviors and complete the final verification score of the output.

[0165] Among them, the inputs include: the protocol structure graph : including conditional acquisition edges; the execution acquisition graph : including path acquisition confidence ; the protocol structure deviation matrix : indicating offset and misaligned structures; the behavior consistency score : providing an abnormal behavior trend score; the causal rule set defined in the protocol , where is the triggering condition, is the indicator to be collected.

[0166] Furthermore, the system judges whether each rule is effectively executed in practice by analyzing the real data of each subject in . That is:

[0167] Whether the triggering condition is met (such as SAE ≥ grade 2, abnormal body temperature, rapid decrease in WBC);

[0168] Whether has been collected within the specified time ;

[0169] Whether the acquisition behavior is highly confident ( high);

[0170] Whether the behavioral trend is reasonable ( high);

[0171] Furthermore, the present invention defines the causal bias score of each subject as follows: ;

[0172] ;

[0173] Where:

[0174] : indicates whether the subject meets the th causal trigger condition (recorded in the analysis , such as the occurrence of an SAE event);

[0175] : the acquisition confidence of this causal path in ;

[0176] : the overall behavioral consistency score of the subject (output of step 4);

[0177] The first item represents that the condition is established but the acquisition is weak or missing; the second item represents that the condition is not established but still acquired, and the behavior is abnormal;

[0178] is the structure - behavior combined weight (recommended value: 0.5 - 0.7);

[0179] is the behavioral abnormality threshold, which is recommended to be set at the 30th percentile of the behavioral score distribution.

[0180] After the system operation is completed, the following results will be output:

[0181] The causal consistency deviation score of each subject ;

[0182] Furthermore, for high - risk subjects, the system will automatically:

[0183] mark the list of paths where the causal trigger fails;

[0184] give reason labels such as "not acquired but should be acquired", "mis - acquired and abnormal behavior", "acquired at the wrong time", etc.;

[0185] push their numbers to the review interface and report generation module for manual review and correction;

[0186] Optionally export the review checklist and Excel report for CRO or regulatory parties to review.

[0187] In one or more embodiments, a clinical trial data verification system based on a modular architecture is disclosed. The system includes:

[0188] A protocol parsing module, configured to obtain a target clinical trial protocol document, generate a protocol structure map according to the time points, collection indicators, and subject populations of the target clinical trial protocol document, and map it into a term mapping table;

[0189] A data fusion module, configured to generate an execution collection map and path confidence based on the protocol structure map and in combination with the actual collected data of the trial;

[0190] A structure verification module, configured to compare the protocol map with the execution collection map, identify structural deviations in the collection path through structural alignment verification, and generate a structural deviation matrix;

[0191] A behavior modeling module, configured to obtain the basic fluctuation range of indicators and medical standards, and generate a consistency behavior score in combination with the generated structural deviation matrix for quantifying the abnormal score of the behavior trend of the subject;

[0192] A causal reasoning module, configured to calculate the causal deviation score of each subject in combination with the protocol structure map, execution collection map, structural deviation matrix, and consistency behavior score, and mark the events of missing collection, wrong collection, and overtime collection in the clinical trial data according to the causal deviation score.

[0193] It should be noted that the specific working process of the clinical trial data verification system based on the modular architecture provided in the embodiments of the present invention is the same as that of the clinical trial data verification method based on the modular architecture described in the above embodiments, and will not be elaborated here.

[0194] The embodiments of the present invention also provide a clinical trial data verification device based on a modular architecture, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps in the embodiments of the clinical trial data verification method based on the modular architecture as described above are implemented, such as Figure 1 the steps S1 - S5 described therein; or, when the processor executes the computer program, the functions of each module in the above system embodiments are implemented.

[0195] Exemplarily, the computer program can be divided into one or more modules. The one or more modules are stored in the memory and executed by the processor to complete the present invention. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the clinical trial data verification device based on the modular architecture.

[0196] The clinical trial data verification device based on a modular architecture may be a computing device such as a desktop computer, a notebook, a palm computer, or a cloud server. The clinical trial data verification device based on a modular architecture may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the clinical trial data verification device based on a modular architecture may further include input / output devices, network access devices, a bus, etc.

[0197] The processor may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The processor is the control center of the clinical trial data verification device based on a modular architecture, and connects various parts of the entire clinical trial data verification device based on a modular architecture through various interfaces and lines.

[0198] The memory may be used to store the computer programs and / or modules. By running or executing the computer programs and / or modules stored in the memory, and by calling the data stored in the memory, the processor implements various functions of the clinical trial data verification device based on a modular architecture. The memory may mainly include a program storage area and a data storage area. Among them, the program storage area may store an operating system, application programs required for at least one function, etc.; the data storage area may store data created according to the operation of the air conditioner controller, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0199] Among them, if the modules integrated in the clinical trial data verification device based on the modular architecture are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-described embodiment methods of the present invention, it can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-described various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0200] Those of ordinary skill in the art can understand that to implement all or part of the processes in the above-described embodiment methods, it can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above-described embodiments of each method. Among them, the storage medium can be a magnetic disk, optical disc, read-only memory (Read-Only Memory, ROM), or random access memory (Random Access Memory, RAM), etc.

[0201] The above is the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art of this technology, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements are also regarded as the protection scope of the present invention.

Claims

1. A method for verifying clinical trial data based on a modular architecture, characterized in that, The method includes: Obtaining a target clinical trial protocol document, generating a protocol structure graph according to the time points, collection indicators, and subject population of the target clinical trial protocol document, and mapping it into a term mapping table; Based on the protocol structure graph, combining the actual collected data of the trial, generating an execution collection graph and path confidence; Comparing the protocol graph with the execution collection graph, and identifying the structural deviations of the collection path through structural alignment verification to generate a structural deviation matrix; Obtaining the basic fluctuation range and medical standards of the indicators, and combining the generated structural deviation matrix to generate a consistency behavior score for quantifying the abnormal score of the subject's behavior trend; Combining the protocol structure graph, execution collection graph, structural deviation matrix, and consistency behavior score, calculating the causal deviation score of each subject, and marking the missed collection, miscollection, and overtime collection events in the clinical trial data according to the causal deviation score.

2. The clinical trial data verification method based on a modular architecture according to claim 1, wherein, The obtaining of the target clinical trial protocol document, generating a protocol structure graph according to the time points, collection indicators, and subject population of the target clinical trial protocol document, and mapping it into a term mapping table specifically includes: Obtaining a target clinical trial protocol document; inputting the target clinical trial protocol document into a BERT-based model, performing structural label prediction on each paragraph, and outputting a structural quadruple, including time points, collection indicators, subject population, and additional conditions in the original protocol that restrict the current collection behavior; Converting the time points, collection indicators, and subject population into three nodes and two edges to form a directed path from time point → subject population → collection indicator; Mapping all nodes to a standard coding system as a term mapping table; where the collection indicators use LOINC; the subject population uses the ICF grouping structure; the time points are converted into the structure of visit days and allowed offset ranges.

3. The clinical trial data verification method based on a modular architecture according to claim 2, wherein The protocol structure graph also has an edge weight regularization mechanism, which specifically includes: Extracting the additional conditions in the original protocol that restrict the current collection behavior from the target clinical trial protocol document, sending the additional conditions in the original protocol that restrict the current collection behavior into the BERT model to obtain the attention weight distribution of each token, calculating its entropy value as a semantic concentration index; and constraining the edge weights according to the semantic concentration index.

4. The clinical trial data verification method based on a modular architecture according to claim 1, wherein Each data of the actual collected data of the trial includes the actual collection time, actual indicator name, and actual subject ID; Then, The generating of the execution collection graph and path confidence based on the protocol structure graph and combining the actual collected data of the trial specifically includes: Performing term standardization processing on the actual indicator name based on the term mapping table to generate an actual standardized term; performing time alignment processing on the actual collection time to obtain the aligned actual time; Based on the results of term standardization and time alignment processing, as well as the actual subject ID, construct an execution acquisition map, and calculate the path confidence; wherein, the execution acquisition map includes several directed acquisition paths, the directed path of the aligned actual time → actual subject ID → actual standardized term; the path confidence is jointly determined by the time matching degree and semantic consistency; the time matching degree is determined based on the actual time node and time point; the semantic consistency is determined by the standardized index node and the acquisition index.

5. The clinical trial data verification method based on a modular architecture according to claim 4, wherein The execution acquisition map is also provided with a local protocol coverage penalty term for controlling the deviation between the number of actual acquisition paths and the number specified by the protocol at a certain time point.

6. The method for verifying clinical trial data based on a modular architecture according to any one of claims 2 or 4, characterized in that Compare the protocol map with the execution acquisition map, and through structural alignment verification, identify the structural deviation of the acquisition path to generate a structural deviation matrix, specifically including: Obtain the paths of the protocol map and the execution acquisition map, and conduct a structural similarity analysis; wherein the structural similarity analysis includes node semantic consistency, population consistency, and time tolerance matching; if the protocol map meets the structural similarity analysis, select the one with the smallest deviation degree as the optimal matching path; Output the minimum deviation degree of all paths of the protocol map and form a structural deviation matrix; Wherein the node semantic consistency means that there is a mapping between the acquisition index and the actual standardized term in the term mapping table, or the cosine similarity ≥ 0.8; the population consistency means that the actual subject ID is the test population; the time tolerance matching means that the actual acquisition time difference does not exceed ±2 days.

7. The method for verifying clinical trial data based on a modular architecture according to claim 6, wherein The deviation degree is calculated as follows: ； Among them, is the deviation degree, which is the confidence level for executing the acquisition spectrum path; is the acquisition index of the protocol spectrum with a local coverage penalty term, indicating the over - acquisition / under - acquisition degree of the acquisition structure; is the offset aggregation regularization term, measuring whether there are a large number of non - standard paths around this time point; and are the regularization term adjustment coefficients.

8. The method for verifying clinical trial data based on a modular architecture according to claim 1, wherein Obtain the basic fluctuation range of the index and the medical standard, combine the generated structural deviation matrix, and generate a consistency behavior score for quantifying the abnormal score of the subject's behavior trend, specifically including: Extract the time series data set of each subject from all the acquisition paths in the execution acquisition map path; Construct the behavior trend tensor of the subject according to the time series data set for depicting the change rate pattern at adjacent time points; Calculating a trend perturbation density regularization term based on the behavior trend tensor and the time series data set , expressed as: ； Among them, T is the total time window, t is the current time window, Indicates The variance of the change rate of all indicators at a time point, is the correlation between the two indicator time series; is the expected correlation of the pair of indicators under normal test population; is the indicator synergistic deviation adjustment factor; Combine the trend perturbation density regularization term and the protocol structure deviation matrix to obtain the final residual term; determine the consistency behavior score through the final residual term.

9. The method for verifying clinical trial data based on a modular architecture according to claim 8, wherein Calculating the causal bias score for each subject , denoted as: ； Among them, indicates whether the subject meets the th causal trigger condition; is the total acquisition window, is the abnormal behavior threshold; indicates the acquisition confidence of the causal path in ; is the structure-behavior joint weight; Among them, represents that the condition is satisfied, but the acquisition is weak or missing; represents that the condition is not satisfied, but the acquisition is still carried out and the behavior is abnormal.

10. A clinical trial data verification system based on a modular architecture, characterized in that, The system includes: A protocol parsing module for obtaining the target clinical trial protocol document, generating a protocol structure map according to the time point, acquisition index, and test population of the target clinical trial protocol document, and mapping it into a term mapping table; A data fusion module for generating an execution acquisition map and path confidence based on the protocol structure map and combining the actual test acquisition data; A structure verification module for comparing the protocol map with the execution acquisition map, and identifying the structural deviation of the acquisition path through structural alignment verification to generate a structural deviation matrix; A behavior modeling module for obtaining the basic fluctuation range of the index and the medical standard, combining the generated structural deviation matrix, and generating a consistency behavior score for quantifying the abnormal score of the subject's behavior trend; A causal inference module, which is used to calculate the causal deviation score of each subject by combining the protocol structure graph, execution collection graph, structure deviation matrix and consistency behavior score, and mark the missed collection, miscollection and overtime collection events in the clinical trial data according to the causal deviation score.

Images