A multi-path causal association arbitration method based on CTCAE grading driving

Abstract

The application provides a multi-path causal association arbitration method based on CTCAE grading driving, relates to the fields of medical information processing and drug warning, and comprises the following steps: acquiring an adverse event candidate event and evidence text thereof, matching the adverse event candidate event with candidate terms in a CTCAE grading table, and corresponding to a common evaluation element table and grading determination rules associated with the terms; extracting the values of observation elements in the common evaluation element table and performing verification, obtaining a degree element according to the element value table, substituting the degree element into a determination expression of the grading determination rules to perform evaluation, and obtaining a severity grading result; inputting the element value table into a plurality of independent algorithm paths respectively to perform parallel operation, and obtaining a causal association determination result or a pending marker; determining a causal association arbitration result according to the causal association determination result or the pending marker output by each algorithm path, and outputting the severity grading result and the causal association arbitration result. The application improves the reproducibility of the severity grading conclusion.

Description

Technical Field

[0001] This invention relates to the fields of medical information processing and pharmacovigilance technology, and in particular to a multi-path causal association arbitration method based on CTCAE hierarchical driving. Background Technology

[0002] In clinical trials, the severity of adverse events is typically graded using grading tables such as CTCAE, while causal relationship assessment relies on a comprehensive evaluation using multiple scale rules or trial protocol-specific rules. This process has long relied primarily on manual interpretation. Different assessors, based on the same evidence, may arrive at inconsistent conclusions due to differences in their understanding of the severity descriptions in the grading tables or their grasp of the scale's assessment pathways. There is room for improvement in assessor consistency. Furthermore, multiple versions of grading tables and related mapping tables are often used concurrently. When a version is updated, it is difficult to retrospectively review previously completed assessments according to the new version's requirements, and it is also difficult to clearly explain the specific table version upon which a particular conclusion was based.

[0003] Chinese invention patent application number 202010313239.X discloses a method, device, terminal, and readable medium for assessing adverse drug reactions. It obtains the assessment score for each adverse event through data field mapping, data extraction, or logical calculation, and then summarizes the results to obtain an overall assessment. However, its analysis strategy is designed for a single assessment process. When multiple independent judgment rules or models process the same adverse event in parallel, and substantial discrepancies arise in the results, it cannot be addressed in a timely manner. Summary of the Invention

[0004] In view of this, the present invention provides a multi-path causal association arbitration method based on CTCAE hierarchical driving, which solves the technical problem in the prior art that there is a lack of unified comparison of the output results of each path when multiple independent judgment rules or models process the same adverse event in parallel, and improves the reproducibility of severity classification conclusions.

[0005] The technical solution of this invention is implemented as follows: This invention provides a multi-path causal association arbitration method based on CTCAE hierarchical driving, comprising: S1. Obtain candidate adverse events and their evidence texts, match the candidate adverse events with candidate terms in the CTCAE grading table, determine the corresponding terms, and retrieve the common evaluation element table and grading judgment rules associated with the corresponding terms. S2. Extract the observation element values ​​from the shared evaluation element table from the evidence text and verify them. Set the values ​​that fail the verification to unknown to obtain the element value table. Combine the observation element values ​​in the element value table according to the synthesis rules recorded in the shared evaluation element table to obtain the severity element values. Substitute the element value table and the severity element values ​​into the judgment expression of the grading judgment rule to evaluate them. Mark each severity level according to whether it has been satisfied, excluded, or not excluded to determine the severity grading result. S3. Input the element value table into multiple independent algorithm paths that do not share intermediate conclusions and perform parallel computation. Each algorithm path corresponds to a different judgment rule system. At least one algorithm path determines the result by weighted summation of the observed element values, and at least one algorithm path determines the result by conditional judgment of each observed element value. If the observed element values ​​involved in the judgment are insufficient to uniquely determine the result, the corresponding algorithm path outputs a pending mark; otherwise, it outputs the causal relationship judgment result. S4. Determine the causal association arbitration result based on the causal association determination results or pending markers output by each algorithm path: When there are pending labels in the output of each algorithm path, the causal association arbitration result is set to be manually reviewed; when there are no pending labels, the causal association judgment results output by each algorithm path are mapped to a unified ordinal scale and grouped and classified. The degree of difference is determined according to whether the classification results are the same. When the degree of difference is zero, it is set to automatic adoption. When the degree of difference is greater than zero and does not exceed the preset threshold, it is set to low confidence adoption. When the degree of difference exceeds the preset threshold, it is set to be manually reviewed. S5. Output the severity grading results and the causal relationship arbitration results.

[0006] Based on the above technical solutions, preferably, step S1 specifically includes: Read the pre-bound configuration version of the adverse event candidate event. The configuration version includes at least the CTCAE classification table version, term mapping table version, thesaurus version, sentence vector model identifier and similarity threshold. Determine the candidate term set in the published CTCAE classification table corresponding to the configuration version. The names of adverse event candidate events are compared with the names of candidate terms in the candidate term set to obtain exact match terms; when there are no exact match terms, the names of adverse event candidate events are compared with the synonyms of candidate terms in the candidate term set to obtain synonym extended match terms. When both exact match terms and synonym expansion terms are empty, semantic similarity is calculated based on the definition and description of the candidate terms in the evidence text and the candidate term set to obtain the semantic similarity of each candidate term; when the maximum value of semantic similarity is lower than the similarity threshold of the configured version, the adverse event candidate event is marked as term cannot be located and the automatic classification ends. When a candidate term exists that is not empty in an exact match, or is not empty in a synonym expansion match, or has a maximum semantic similarity of not less than the similarity threshold, the candidate term is identified as the corresponding term. The common evaluation element table associated with the corresponding term in the published CTCAE grading table and the grading determination rules are retrieved. The common evaluation element table includes observation elements, degree elements and the synthesis rules of degree elements. The grading determination rules record the determination expressions for each severity level.

[0007] Based on the above technical solutions, preferably, step S2 specifically includes: S21. Determine the description judgment results and evidence location markers of each observation element from the evidence text according to the preset set of acceptable description forms. S22. Verify the description judgment result and evidence location identifier with the evidence text. If the verification fails, set the value of the observed element to unknown. If the verification passes, write the value indicated by the description judgment result into the observed element to obtain the element value table. S23. According to the synthesis rules recorded in the common evaluation element table, the degree element value is obtained from the observed element value in the element value table; S24. Substitute the element value table and the degree element value into the judgment expression recorded in the grading judgment rule to obtain the severity grading result.

[0008] Based on the above technical solutions, preferably, step S22 specifically includes: Read the text fragment pointed to by the evidence location marker, determine whether the text fragment falls within the scope of the selected description form or the registered synonyms of the selected description form in the description judgment result, and obtain the location consistency judgment result; If the location consistency determination result is inconsistent, or the description determination result does not belong to the preset set of acceptable description forms, or the content indicated by the description determination result is inconsistent with the text fragment pointed to by the evidence location marker, the verification is determined to be unsuccessful and the value of the observed element is set to unknown; otherwise, the verification is determined to be successful and the value indicated by the description determination result is written into the element value table. For numerical observation features, the verification method is determined based on the type of data source: For laboratory numerical observation elements derived from structured test reports, the test report values ​​are directly written into the element value table and the source location of the values ​​is retained. For laboratory numerical observation elements extracted from the evidence text, an evidence location identifier is given, and the description judgment result and the evidence location identifier are verified with the evidence text. If the verification fails, the laboratory numerical value is set to unknown.

[0009] Based on the above technical solutions, preferably, step S24 specifically includes: Substitute the numerical observation elements and degree elements from the element value table into the judgment expression recorded in the grading judgment rule, and mark them as satisfied, excluded, or not yet excluded due to unknown values ​​according to each severity level. The highest-ranking severity level that is marked as satisfied is... The highest-ranking severity level among those marked as not yet ruled out is... ,in This indicates the highest severity level of the status that has been met. This indicates the highest severity level among the states that have not yet been ruled out; when it is higher than... When all possible severity levels have been excluded, the severity classification result will be taken as a point level. When higher than If any severity level has not yet been ruled out, the severity classification result will be taken as an interval. When there are no satisfied levels but there are levels that have not yet been excluded, the severity classification result is represented by an upper bound of ; The interval with an undetermined lower bound.

[0010] Based on the above technical solutions, the preferred one is... The multiple algorithm paths are specifically a first independent algorithm path, a second independent algorithm path, and a third independent algorithm path; step S3 specifically includes: S31. Assign scores to the observed element values ​​in the element value table according to the preset weighted scoring rules and sum them to obtain the weighted total score S. Determine the causal relationship judgment result or pending label of the first independent algorithm path based on the preset score range of S. S32. According to the preset multi-dimensional condition judgment rules, judge the values ​​of the observed elements in the element value table one by one, and determine the causal relationship judgment result or pending mark of the second independent algorithm path. S33. Match the values ​​of observed elements in the element value table according to the rule chain formed by the correspondence between conditions and conclusions. The correspondence between conditions and conclusions is compiled from the clinical trial protocol. Determine the causal association judgment result or pending label of the third independent algorithm path according to the triggered correspondence between conditions and conclusions.

[0011] Based on the above technical solutions, preferably, the parallel computation specifically includes: When there are observation elements with unknown values ​​in the first independent algorithm path, for each observation element with an unknown value, the values ​​included in the preset allowed value range are substituted one by one into the weighted scoring rule to calculate the total score, resulting in multiple total scores; the minimum total score among the multiple total scores is recorded as . The maximum total score is recorded as To obtain the total score range When the total score interval crosses the correlation level boundary corresponding to the preset score interval, the output of the first independent algorithm path is set to the pending mark; when the total score interval falls within the same correlation level, the output of the first independent algorithm path is set to the corresponding causal correlation category. When the value of the observed element participating in the judgment in the second independent algorithm path is unknown, or the judgment path falls into the preset multi-level condition branches and is marked as a branch that cannot be classified or evaluated, the output of the second independent algorithm path is set to the pending label; otherwise, it is set to the causal association category corresponding to the judgment path. When the observation element required by the rule chain in the third independent algorithm path is missing or has an unknown value in the element value table, or when the rule chain determines that the data is insufficient, the output of the third independent algorithm path is set to the pending label; otherwise, it is set to the causal association category triggered by the rule chain.

[0012] Based on the above technical solutions, preferably, the step of mapping the causal association determination results output by each algorithm path to a unified ordinal scale specifically involves: mapping according to a causal association determination result and ordinal scale value lookup table bound to the configuration version, and classifying the mapped ordinal scale values ​​according to preset association level groups. The configuration version is a version identifier pre-bound to the candidate adverse event, and the lookup table remains unchanged after the configuration version is released. When the association level groups to which each algorithm path belongs are the same, the degree of difference is recorded as zero. When the association level groups to which each algorithm path belongs are different, the difference between the maximum and minimum values ​​of the ordinal scale values ​​output by each algorithm path is taken as the degree of difference.

[0013] Based on the above technical solutions, preferably, the step S1 includes the following: A consistency check is performed on the degree elements and the judgment expressions in the grading judgment rules recorded in the common evaluation element table. The following situations are marked as defects, and the consistency check results are obtained: degree elements that hit the preset degree vocabulary but are not included in the exemption list; judgment expressions that cannot be satisfied; degree elements that have been defined but have never been referenced by any judgment expression. When the consistency verification result does not contain any defects, the shared evaluation element table, the grading judgment rules, the preset set of acceptable description forms, and the list of exemption items will be published as the published version; if it contains defects, it will not be published. To bind a configuration version to an adverse event candidate, if there is no corresponding published version for the configuration version, the adverse event candidate must not be matched with the CTCAE classification table.

[0014] Further preferred options include: The element value table, evidence location identifier, degree element value, severity classification result, corresponding terminology, ordinal scale value, degree of difference, causal relationship arbitration result, and the bound configuration version, causal relationship determination result or pending mark of each independent algorithm path are written to the audit log in an append-only manner. Records already written to the audit log must not be overwritten. When the configuration version changes, steps S1 to S4 are re-executed for the processed adverse event candidate events bound to the configuration version involved in the change, based on the changed configuration version, to obtain the recalculation results. An impact list is generated that records the changes in the conclusions. The impact list includes adverse event candidate events, the severity classification results or causal relationship arbitration results before and after the change, and the reasons for the change. The recalculation results and the impact list are written to the audit log in an append-only manner. When it is necessary to retain two conclusions before and after the change, a new event version record is created and the old record is frozen. The record is marked as a cross-version related record in the audit log.

[0015] The present invention has the following advantages over the prior art: (1) The verification of observation elements, the synthesis of degree elements and the evaluation process of severity classification are uniformly incorporated into the common evaluation element table and classification judgment rules for processing, so that the classification conclusion has a definite correspondence under the condition that the input facts and the judgment rules are determined, thereby improving the reproducibility of severity classification conclusion; (2) Multiple independent algorithm paths with different judgment rule systems are used to complete the causal relationship judgment in parallel. Each path outputs a pending mark instead of giving a conclusion when the value of the element involved in the judgment is incomplete, thus avoiding the generation of judgment results without basis in the case of insufficient information. (3) The causal relationship judgment results output by each path are mapped to a unified ordinal scale and grouped and classified. Manual review is triggered only when there are still substantial differences after grouping. This avoids the superficial differences caused by inconsistent classification of different judgment rule systems being misjudged as substantial differences, and also avoids the simple classification of pending labels into a certain degree of association. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a multi-path causal association arbitration method based on CTCAE hierarchical driving according to the present invention. Figure 2 This is a system framework diagram of a multi-path causal association arbitration method based on CTCAE hierarchical driving according to the present invention; Figure 3 This is a flowchart of a consistency verification method for a multi-path causal association arbitration method based on CTCAE hierarchical driving according to the present invention. Figure 4 This is a flowchart of a term matching process for a multi-path causal association arbitration method based on CTCAE hierarchical driving according to the present invention. Figure 5 This is a schematic diagram illustrating the severity classification of a multi-path causal association arbitration method based on CTCAE hierarchical driving according to the present invention. Figure 6 This is a schematic diagram of multi-path parallel computation of a multi-path causal association arbitration method based on CTCAE hierarchical driving according to the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figure 1 As shown, this invention provides a multi-path causal association arbitration method based on CTCAE hierarchical driving, comprising: S1. Obtain candidate adverse events and their evidence texts, match the candidate adverse events with candidate terms in the CTCAE grading table, determine the corresponding terms, and retrieve the common evaluation element table and grading judgment rules associated with the corresponding terms.

[0020] Understandably, CTCAE (Common Adverse Event Evaluation Criteria) is a standardized terminology system developed by the National Cancer Institute (NCI) for assessing adverse events (AEs) observed during cancer treatment. It categorizes adverse events into 1-5 levels, with higher levels indicating more severe symptoms. This invention addresses already obtained candidate adverse events, which must include at least the original terminology, clinical description, laboratory results, and a timeline of medication use and outcome. A shared evaluation element table is stored in a structured data table, organized by CTCAE terminology. Basic fields include at least system organ classification, AE terminology name, severity level, level definition, laboratory threshold field, and clinical criterion field. Each severity level under the same AE terminology no longer lists its own independent criteria but shares the same evaluation element table. The clinical criterion field for each level is simply a judgment expression on this shared table. This allows for the reuse and comparison of criteria between different levels under the same term, avoiding redundancy and inconsistencies caused by unrelated criteria across levels.

[0021] This invention binds the processing of adverse event candidate events to a specific configuration version, and performs a consistency verification mechanism on the grading criteria, judgment rules and related entries before release, thereby improving the traceability of the standards on which the evaluation process depends when the version changes.

[0022] like Figure 2 As shown, the processing flow of this invention can be achieved by the candidate event and evidence text acquisition module of the input layer obtaining the original data, and then sequentially passing through the terminology matching module, element extraction and verification module, severity classification module, causal relationship determination module, arbitration and difference calculation module of the processing layer to complete the processing; the shared evaluation element table library, classification judgment rule library, consistency verification unit and version release unit of the evaluation element management layer provide the processing layer with the published judgment data; the manual confirmation and review interaction module of the interaction and configuration layer undertakes the manual intervention link when the terminology cannot be located, the conclusion has a pending mark, or the difference degree exceeds the threshold; the configuration version management module is responsible for binding the configuration version to the candidate event; the audit log storage module of the storage layer records the key events in the processing process in an append-only manner.

[0023] like Figure 3 and Figure 4 As shown, specifically, step S1 includes: Read the pre-bound configuration version of the adverse event candidate event. The configuration version includes at least the CTCAE classification table version, term mapping table version, thesaurus version, sentence vector model identifier and similarity threshold. Determine the candidate term set in the published CTCAE classification table corresponding to the configuration version. The names of adverse event candidate events are compared with the names of candidate terms in the candidate term set to obtain exact match terms; when there are no exact match terms, the names of adverse event candidate events are compared with the synonyms of candidate terms in the candidate term set to obtain synonym extended match terms. When both exact match terms and synonym expansion terms are empty, semantic similarity is calculated based on the definition and description of the candidate terms in the evidence text and the candidate term set to obtain the semantic similarity of each candidate term; when the maximum value of semantic similarity is lower than the similarity threshold of the configured version, the adverse event candidate event is marked as term cannot be located and the automatic classification ends. When a candidate term exists that is not empty in an exact match, or is not empty in a synonym expansion match, or has a maximum semantic similarity of not less than the similarity threshold, the candidate term is identified as the corresponding term. The common evaluation element table associated with the corresponding term in the published CTCAE grading table and the grading determination rules are retrieved. The common evaluation element table includes observation elements, degree elements and the synthesis rules of degree elements. The grading determination rules record the determination expressions for each severity level.

[0024] Understandably, the elements recorded in the shared evaluation element table are divided into two categories: observation elements and degree elements. Among them, the observation elements include (1) severity-related facts obtained by splitting the original text of the CTCAE terms corresponding to the candidate adverse events, which are used for severity classification; (2) the relevant facts extracted from the same evidence text required for causal relationship determination, including at least the temporal relationship between medication and the occurrence of adverse events, de-stimulation reaction (whether the event is relieved after stopping or reducing the dosage), re-stimulation reaction (whether the event recurs after medication is administered again), whether alternative causes that can explain the event can be excluded, previous reports of similar adverse events, and supporting laboratory or auxiliary examination evidence.

[0025] By splitting the criteria into two layers—observational elements and degree elements—the information extraction process of the evidence text only requires the selection of a closed set at the factual level, eliminating the need for subjective judgment on degree-related or evaluative statements. The values ​​of the degree elements are calculated entirely from the observational elements by the published synthesis rules, thereby limiting the subjective judgment process to outside the evidence extraction stage and improving the reproducibility of the processing results under the same input.

[0026] The term matching process proceeds sequentially through three levels: exact matching, synonym expansion, and semantic similarity. If a term is matched at the previous level, the process does not proceed to the next level. Semantic similarity is only calculated when no term is matched at the first two levels. This avoids errors in retrieving subsequent evaluation element tables due to mismatches of semantically similar but actually different terms.

[0027] In one embodiment of the present invention, the semantic similarity can be calculated by using a pre-trained language model to encode the evidence text of the candidate event and the definition description of the candidate term into vectors, and then calculating the cosine similarity. The present invention does not impose any specific limitations on this.

[0028] In one embodiment of the present invention, the method further includes the following step before step S1: A consistency check is performed on the degree elements and the judgment expressions in the grading judgment rules recorded in the common evaluation element table. The following situations are marked as defects, and the consistency check results are obtained: degree elements that hit the preset degree vocabulary but are not included in the exemption list; judgment expressions that cannot be satisfied; degree elements that have been defined but have never been referenced by any judgment expression. When the consistency verification result does not contain any defects, the shared evaluation element table, the grading judgment rules, the preset set of acceptable description forms, and the list of exemption items will be published as the published version; if it contains defects, it will not be published. To bind a configuration version to an adverse event candidate, if there is no corresponding published version for the configuration version, the adverse event candidate must not be matched with the CTCAE classification table.

[0029] Understandably, the judgment expression is a set of evaluable conditions recorded in the grading judgment rules for each CTCAE severity level, and it mainly falls into two categories: Numerical comparison: Compare the observed data with laboratory thresholds, such as whether the test value exceeds a certain upper limit; Truth condition: Use the synthesized degree elements to determine whether they are true or false, such as whether daily life is restricted; The judgment expression contains only two types: one is numerical comparison, such as whether the test value exceeds a certain upper limit; the other is the truth condition for the degree element value, such as whether daily life is restricted. After substituting the observed element value and the degree element value into the judgment expression, the expression will only produce three results: satisfied, excluded, and not excluded.

[0030] Among them, the consistency of the observation elements, degree elements and the judgment expressions in the grading judgment rules recorded in the common evaluation element table is checked, and the following situations are marked as defective items: (1) An element is labeled as an observation element, but its name or definition matches the preset degree term list and is not included in the exemption list. This situation indicates that the element actually has a degree meaning but is incorrectly labeled as an observation element. For example, the exemption list can register elements such as "death" and "immediate life-threatening intervention" which literally contain degree terms, but are used as fixed threshold judgment items in the CTCAE original text. (2) If there is no possible combination of assignments that makes a certain judgment expression true within the possible value range of the elements it references, then the expression is unsatisfactory. (3) A certain observation element or degree element has been defined in the common evaluation element table, but has not been referenced by any judgment expression or synthesis rule, that is, the element is redundant.

[0031] The evaluation element table can only be published and registered if none of the above-mentioned defect items are present in the shared evaluation element table; the candidate event processing flow can only begin after the evaluation element table is published.

[0032] Furthermore, when neither exact matching nor synonym expansion matching matches the corresponding term, and the maximum semantic similarity is lower than the similarity threshold of the configured version, the candidate event for the adverse event is marked as a term that cannot be located, written to the audit log in an append-only manner, and transferred to the manual term confirmation stage; after the corresponding term is manually confirmed, the subsequent process continues from the step of retrieving the common evaluation element table to which the term belongs.

[0033] When multiple candidate terms are matched by exact matching or synonym expansion matching, the term with the highest preset priority in the match set is selected; when the maximum value in semantic similarity matching corresponds to a unique term, that term is selected; when the maximum value corresponds to multiple terms, it is also marked as a term that cannot be located and the process is transferred to the manual term confirmation stage.

[0034] S2. Extract the observation element values ​​from the shared evaluation element table from the evidence text and verify them. Set the values ​​that fail the verification to unknown to obtain the element value table. Combine the observation element values ​​in the element value table according to the synthesis rules recorded in the shared evaluation element table to obtain the severity element values. Substitute the element value table and the severity element values ​​into the judgment expression of the grading judgment rule to evaluate them. Mark each severity level according to whether it has been satisfied, excluded, or not excluded to determine the severity grading result. like Figure 5 As shown, specifically, step S2 includes: S21. Determine the description judgment results and evidence location markers of each observation element from the evidence text according to the preset set of acceptable description forms. S22. Verify the description judgment result and evidence location identifier with the evidence text. If the verification fails, set the value of the observed element to unknown. If the verification passes, write the value indicated by the description judgment result into the observed element to obtain the element value table. S23. According to the synthesis rules recorded in the common evaluation element table, the degree element value is obtained from the observed element value in the element value table; S24. Substitute the element value table and the degree element value into the judgment expression recorded in the grading judgment rule to obtain the severity grading result.

[0035] The process of determining the descriptive judgment results of each observation element from the evidence text is accomplished using conventional information extraction methods or manual annotation. This invention does not impose specific restrictions on this, only requiring that the output be limited to a preset set of acceptable descriptive forms and that an evidence location identifier be provided. The preset set of acceptable descriptive forms is a limited set of descriptive forms pre-defined for each observation element. For example, for a certain observation element, this set can be set to three forms: recorded, unrecorded, and undeterminable. When determining the value of an observation element from the evidence text, only one descriptive form can be selected from this set as the descriptive judgment result; freely generated expressions outside the set are not allowed. Simultaneously, the location information of the descriptive judgment result in the evidence text must be provided as an evidence location identifier. After verification, the observation element is written into the element value table. Then, the published synthesis rules for that term in the shared evaluation element table are read, and the degree element is calculated using the verified observation element. The synthesis rules are published along with the shared evaluation element table. Only verified observation elements are read, and synthesis is performed according to the priority order stipulated in the rules to obtain a unique value. If there are unknowns among the observation elements participating in the synthesis, it is first determined whether the unknowns will change the synthesis result. Regardless of the value of the unknown, if the result is the same, then that determined value will still be output. If different values ​​of the unknown will cause changes in the result, then that degree element will be marked as unknown and will be processed as not yet excluded. When information is incomplete, degree elements will not be automatically grouped into a specific category.

[0036] This invention integrates the verification of observation elements, the synthesis of degree elements, and the evaluation of severity grading into a common evaluation element table and grading rules, so that the grading conclusions have a definite correspondence under the condition that the input facts and the criterion rules are determined, thereby improving the reproducibility of severity grading conclusions.

[0037] In one embodiment of the present invention, step S22 specifically includes: Read the text fragment pointed to by the evidence location marker, determine whether the text fragment falls within the scope of the selected description form or the registered synonyms of the selected description form in the description judgment result, and obtain the location consistency judgment result; If the location consistency determination result is inconsistent, or the description determination result does not belong to the preset set of acceptable description forms, or the content indicated by the description determination result is inconsistent with the text fragment pointed to by the evidence location marker, the verification is determined to be unsuccessful and the value of the observed element is set to unknown; otherwise, the verification is determined to be successful and the value indicated by the description determination result is written into the element value table. For numerical observation features, the verification method is determined based on the type of data source: For laboratory numerical observation elements derived from structured test reports, the test report values ​​are directly written into the element value table and the source location of the values ​​is retained. For laboratory numerical observation elements extracted from the evidence text, an evidence location identifier is given, and the description judgment result and the evidence location identifier are verified with the evidence text. If the verification fails, the laboratory numerical value is set to unknown.

[0038] If the evidence location marker is incorrectly located, the description judgment result exceeds the preset set, or the description judgment result is inconsistent with the content pointed to by the evidence location marker it references, the verification will fail if any of these situations occur. The value of the observed element will be set to unknown instead of the original output. This ensures that the final value of each observed element is cross-verified with the evidence text, and the source of the value can be traced back to the specific location in the evidence text. This avoids obtaining an unverified definite value when the evidence is insufficient or the extraction is incorrect.

[0039] In one embodiment of the present invention, the degree element is calculated based on the observed element values ​​in the element value table according to the synthesis rules recorded in the shared evaluation element table. The synthesis rules themselves are published along with the shared evaluation element table. Taking a degree element representing the degree of limitation in daily living activities as an example, its synthesis rules can be jointly determined by several observed elements related to self-care ability and several observed elements related to instrumental daily living ability: when there are items with limited values ​​among the observed elements related to self-care ability, the degree element is synthesized as self-care limited; when none of the observed elements related to self-care ability are limited, but there are items with limited values ​​among the observed elements related to instrumental daily living ability, it is synthesized as instrumental limited; when neither type of observed element has items with limited values, it is synthesized as unrestricted. This synthesis rule determines a unique degree element value by combining the values ​​of several observed elements according to a pre-agreed priority order. During operation, only the verified observed element values ​​need to be read and substituted into the synthesis rule to obtain the determined synthesis result. Different specific synthesis rules can be used for degree elements corresponding to different terms, all of which are recorded in the shared evaluation element table.

[0040] When a certain level element depends on an observation element with an unknown value, it is handled according to the following rules: if the composite result remains consistent regardless of the value assigned to the unknown item within its allowed value range, then the determined composite result is still output; if different values ​​assigned to the unknown item would cause the composite result to change, then the value of the level element is set to unknown and processed as a level that has not yet been excluded, and it is not directly composited into an unrestricted or a certain restricted type.

[0041] In one embodiment of the present invention, step S24 specifically includes: Substitute the numerical observation elements and degree elements from the element value table into the judgment expression recorded in the grading judgment rule, and mark them as satisfied, excluded, or not yet excluded due to unknown values ​​according to each severity level. The highest-ranking severity level that is marked as satisfied is... The highest-ranking severity level among those marked as not yet ruled out is... ,in This indicates the highest severity level of the status that has been met. This indicates the highest severity level among the states that have not yet been ruled out; when it is higher than... When all possible severity levels have been excluded, the severity classification result will be taken as a point level. When higher than If any severity level has not yet been ruled out, the severity classification result will be taken as an interval. When there are no satisfied levels but there are levels that have not yet been excluded, the severity classification result is represented by an upper bound of ; An interval with an undetermined lower bound. When there are no grades that meet the criteria and no grades that have not yet been excluded, the severity classification result is recorded as unclassifiable. Interval This indicates the lower bound that has been satisfied and the upper bound that has not yet been excluded, but does not mean that every intermediate level within this interval has not been excluded.

[0042] This invention labels the judgment expressions for each severity level as three states: satisfied, excluded, and not excluded. Only expressions above a certain severity level are considered satisfied. The point level is output only when all levels have been explicitly excluded; if a higher level exists that cannot be excluded due to unknown values, the overlay is output. to The interval results accurately reflect the completeness of the evidence and will not be misjudged as lower level due to missing information in individual elements. The numerical observation elements in the judgment expression are compared with the laboratory threshold in the same way, while the degree elements are judged by substituting their composite results into the truth conditions of the judgment expression. Both use the same set of marking rules for satisfied, excluded, and not yet excluded, and do not use different methods to handle missing information due to different sources of judgment criteria.

[0043] In one embodiment of the present invention, step S24 further includes: comparing the values ​​of the observed elements in the element value table with the preset severe adverse event determination criteria item by item to obtain the hit items that match the determination conditions and the evidence location identifier of the observed elements corresponding to the hit items. The comparison uses an element value table, and the severe adverse event determination criteria record multiple preset classification items for the consequences of adverse events. Specific classification items can be set with reference to the generally accepted severe adverse event determination criteria in the field of pharmacovigilance. This comparison and the determination of the severity grading results are independent of each other; the same candidate event can hit the severe adverse event determination conditions at a lower severity level, and the two do not constitute a mutually exclusive relationship.

[0044] S3. Input the element value table into multiple independent algorithm paths that do not share intermediate conclusions and perform parallel computation. Each algorithm path corresponds to a different judgment rule system. At least one algorithm path determines the result by weighted summation of the observed element values, and at least one algorithm path determines the result by conditional judgment of each observed element value. If the observed element values ​​involved in the judgment are insufficient to uniquely determine the result, the corresponding algorithm path outputs a pending mark; otherwise, it outputs the causal relationship judgment result. like Figure 6 As shown, specifically, the multiple algorithm paths are a first independent algorithm path, a second independent algorithm path, and a third independent algorithm path; step S3 specifically includes: S31. Assign scores to the observed element values ​​in the element value table according to the preset weighted scoring rules and sum them to obtain the weighted total score S. Determine the causal relationship judgment result or pending label of the first independent algorithm path based on the preset score range of S. S32. According to the preset multi-dimensional condition judgment rules, judge the values ​​of the observed elements in the element value table one by one, and determine the causal relationship judgment result or pending mark of the second independent algorithm path. S33. Match the values ​​of observed elements in the element value table according to the rule chain formed by the correspondence between conditions and conclusions. The correspondence between conditions and conclusions is compiled from the clinical trial protocol. Determine the causal association judgment result or pending label of the third independent algorithm path according to the triggered correspondence between conditions and conclusions.

[0045] This invention employs multiple independent algorithm paths, each corresponding to a different judgment rule system, to complete the causal association determination in parallel. When the values ​​of the elements involved in the judgment are incomplete, each path outputs a pending mark instead of giving a conclusion reluctantly, thus avoiding judgment results that lack evidence in the case of insufficient information.

[0046] In one embodiment of the present invention, the weighted scoring rule based on the first independent algorithm path can adopt the questions and scoring rules of the Naranjo scale. Each observed element is scored and summed, and the association level is output according to the scale's conventional classification: certain, very likely, possible, and doubtful. Unlike the original Naranjo scale rules, which assign 0 points to uncertain scoring items, in this path, when the value of an observed element corresponding to a scoring item is unknown, and any allowed value for that item could cause the total score interval to cross the preset score boundary, this path outputs a pending mark and does not forcibly give a certain level.

[0047] Record No. Each observed element is assigned a score according to a preset weighted scoring rule. (If the value of this element is unknown, it will not be pre-assigned 0 points, but will be allowed to be substituted one by one), weighted total score. ; When using the Naranjo scale, Take the score of the scale for the corresponding question.

[0048] In one embodiment of the present invention, the decision order and multi-level conditional branches upon which the second independent algorithm path is based can be evaluated using the WHO-UMC causal relationship assessment decision matrix. Specifically, the conditions are judged sequentially according to the following five dimensions: First, whether the time sequence and interval between medication and adverse event occurrence conform to a reasonable temporal correlation; second, whether the adverse event is alleviated after medication is discontinued or dosage reduced, i.e., de-stimulation response; third, whether the adverse event recurs after medication is re-administered, i.e., re-stimulation response; fourth, whether the adverse event can be sufficiently explained by concomitant medication, primary disease, or other known causes, i.e., the degree of exclusion of alternative causes; and fifth, whether there are laboratory tests or other objective evidence supporting the association. The judgment results of each dimension are classified according to preset combination rules, with categories ranging from high to low as: certain, very likely, possible, unlikely, to be evaluated, and cannot be evaluated.

[0049] In one embodiment of the present invention, the rule chain on which the third independent algorithm path is based is compiled from the drug-specific association rules recorded in the clinical trial safety monitoring plan. The compilation is into priority production rules. During operation, the preconditions are judged one by one in priority order. The first rule that satisfies the preconditions outputs its corresponding conclusion. If the preconditions of any rule are not satisfied after traversing the rule chain in the above order, the rule chain is judged to be insufficient data.

[0050] The rule chain is configured as follows: Text in clinical trial protocols or product instructions regarding correlation judgments such as post-discontinuation outcomes, reactivation, and concomitant medication is organized into several "prerequisite-conclusion" rules, arranged in a preset priority order. During runtime, the corresponding observation elements in the element value table are read sequentially. If all prerequisites are met, the conclusion for that rule is triggered, and subsequent rules are stopped. If the required observation element is missing, unknown, or none of the rules are triggered, it is considered insufficient data. For example: Rule 1 is "If the adverse event is relieved after discontinuation and recurs after reactivation, it is considered highly probable"; Rule 2 is "If there are other clear etiological explanations, it is considered possibly irrelevant"; Rule 3 is "If the data is insufficient to support any of the above prerequisites, it is considered insufficient data." This invention does not limit the number of rules or their specific wording; they only need to be pre-registered in the configuration version and can be read by a third independent algorithm path.

[0051] In one embodiment of the present invention, parallel computation specifically includes: When there are observation elements with unknown values ​​in the first independent algorithm path, for each observation element with an unknown value, the values ​​included in the preset allowed value range are substituted one by one into the weighted scoring rule to calculate the total score, resulting in multiple total scores; the minimum total score among the multiple total scores is recorded as . The maximum total score is recorded as To obtain the total score range When the total score interval crosses the correlation level boundary corresponding to the preset score interval, the output of the first independent algorithm path is set to the pending mark; when the total score interval falls within the same correlation level, the output of the first independent algorithm path is set to the corresponding causal correlation category. When the value of the observed element participating in the judgment in the second independent algorithm path is unknown, or the judgment path falls into the preset multi-level condition branches and is marked as a branch that cannot be classified or evaluated, the output of the second independent algorithm path is set to the pending label; otherwise, it is set to the causal association category corresponding to the judgment path. When the observation element required by the rule chain in the third independent algorithm path is missing or has an unknown value in the element value table, or when the rule chain determines that there is insufficient data (i.e., no rule is triggered after judging each precondition), the output of the third independent algorithm path is set to the pending mark; otherwise, it is set to the causal association category triggered by the rule chain.

[0052] All three independent algorithm paths follow the same principle: when the values ​​of the factors involved in the judgment are insufficient to support a definite conclusion, a pending label is output, rather than forcibly selecting a specific causal relationship category. Taking the first independent algorithm path as an example, the impact of unknowns on the total score is not simply zeroed out or ignored. Instead, a total score interval is calculated by substituting the values ​​that minimize and maximize the total score. Only when this interval falls entirely within the score range corresponding to the same level will the existence of unknowns not affect the final level assignment, and in this case, the level can be output. Once the interval crosses the level boundary, it indicates that the value of the unknowns is sufficient to influence the final level assignment, and in this case, no specific level should be output, but a pending label should be output. The pending judgment principle in the second and third independent algorithm paths is the same: when the factor information is insufficient during the judgment process, and the judgment path or rule chain cannot be uniquely determined, a pending label is output, rather than forcibly giving a conclusion based on incomplete information.

[0053] S4. Determine the causal association arbitration result based on the causal association determination results or pending markers output by each algorithm path: When there are pending labels in the output of each algorithm path, the causal association arbitration result is set to be manually reviewed; when there are no pending labels, the causal association judgment results output by each algorithm path are mapped to a unified ordinal scale and grouped and classified. The degree of difference is determined according to whether the classification results are the same. When the degree of difference is zero, it is set to automatic adoption. When the degree of difference is greater than zero and does not exceed the preset threshold, it is set to low confidence adoption. When the degree of difference exceeds the preset threshold, it is set to be manually reviewed.

[0054] Specifically, mapping the causal association determination results output by each algorithm path to a unified ordinal scale involves: mapping according to a lookup table of causal association determination results and ordinal scale values ​​bound to the configuration version; and classifying the mapped ordinal scale values ​​into preset association level groups. The configuration version is a version identifier pre-bound to the candidate adverse event, and the lookup table remains unchanged after the configuration version is released. When the association level groups to which each algorithm path belongs are the same, the degree of difference is recorded as zero. When the association level groups to which each algorithm path belongs are different, the difference between the maximum and minimum values ​​of the ordinal scale values ​​output by each algorithm path is taken as the degree of difference.

[0055] S5. Output the severity grading results and the causal relationship arbitration results.

[0056] In one embodiment of the present invention, step S4 specifically includes: Obtain the causal relationship determination results or pending labels output by the first independent algorithm path, the second independent algorithm path, and the third independent algorithm path, respectively; According to the causal association determination results and ordinal scale value comparison table bound to the configuration version, the causal association determination results output by the first independent algorithm path, the second independent algorithm path, and the third independent algorithm path are mapped to ordinal scale values ​​respectively, and the association level group to which the first independent algorithm path, the second independent algorithm path, and the third independent algorithm path belong is determined according to the ordinal scale values; the causal association determination results and ordinal scale value comparison table remains unchanged after the configuration version is released; When the association level groups to which the first, second, and third independent algorithm paths belong are the same, the degree of difference is recorded as zero; when the association level groups to which the first, second, and third independent algorithm paths belong are different, the difference between the maximum and minimum values ​​of the ordinal scale values ​​output by the first, second, and third independent algorithm paths is taken as the degree of difference. When the degree of difference is zero, the causal association arbitration result is set to automatic adoption; when the degree of difference is greater than zero but does not exceed the preset threshold, the causal association arbitration result is set to low confidence adoption; when the degree of difference exceeds the preset threshold, or when there is a pending marker in the causal association determination result output by the first independent algorithm path, the second independent algorithm path, and the third independent algorithm path, the causal association arbitration result is set to be transferred to manual review.

[0057] Understandably, when the output of any independent algorithm path is a pending label, no ordinal scale mapping is performed on the output of the path, and the causal association arbitration result is directly set to be manually reviewed; only when each independent algorithm path outputs a specific causal association determination result will the following ordinal scale mapping and difference degree calculation be performed.

[0058] This invention, after mapping the output results of each path to a unified ordinal scale value, further divides the ordinal scale values ​​into high-association and low-association groups according to a preset boundary value. Differences in ordinal scale values ​​within the same group are recorded as zero; only when they belong to different groups are the degree of difference calculated. This filters out superficial differences caused by variations in the coarseness of the judgment rule system, retaining only substantial disagreements across association level groups as the basis for calculating the degree of difference. A degree of difference of zero indicates that the output results of each path are consistent or only have superficial differences within the same group; in this case, the causal association arbitration result can be automatically adopted. A degree of difference greater than zero but not exceeding a preset threshold indicates that there is a certain degree of substantial disagreement between the paths, but the disagreement is still within an acceptable range; in this case, it is set to low confidence adoption to distinguish it from completely consistent automatically adopted results. This avoids superficial disagreements caused by inconsistent coarseness of different judgment rule systems being misjudged as substantial disagreements, and also avoids simply classifying pending labels into a certain degree of association. When automatically adopting or adopting with low confidence, the association category in the causal association arbitration result is determined according to the following rules: If a main path is specified in advance, the causal association judgment result output by the main path is used as the association category; if no main path is specified, the causal association judgment result corresponding to the largest ordinal scale value is used as the association category. When adopting with low confidence, the association category is still given according to the above rules, and the low confidence level is also marked.

[0059] In one embodiment of the present invention, the ordinal scale value is 1, 2, 3, 4, or 5. Results with an ordinal scale value of 4 or 5 are grouped into a high-association group, results with an ordinal scale value of 1 or 2 are grouped into a low-association group, and results with an ordinal scale value of 3 are grouped separately. When all ordinal scale values ​​belong to the same association level group, the degree of difference is... The value is set to zero; when the ordinal scale values ​​belong to different association level groups, the maximum value among the ordinal scale values ​​belonging to different association level groups is recorded as zero. The minimum value is ,according to Calculate the degree of difference ;in This represents the maximum value among the ordinal scale values ​​belonging to different association level groups. This represents the minimum ordinal scale value among groups belonging to different association levels. This indicates the degree of difference after alignment. For example, the lookup table frozen with the configuration version can be set as follows: "Confirm" output by the first independent algorithm path, "Affirmative" output by the second independent algorithm path, and the highest-level conclusion triggered by the third independent algorithm path are all mapped to an ordinal scale value of 5; "Very likely" is mapped to 4; "Possible" is mapped to 3; "Unlikely" or "Suspicious" is mapped to 2; "To be evaluated" or "Cannot be evaluated" is mapped to 1. The specific correspondence is recorded with the configuration version and can be adjusted accordingly, without affecting the arbitration logic after unified mapping due to the different classification granularity of the original conclusions of the three independent algorithm paths.

[0060] In one embodiment of the present invention, a multi-path causal association arbitration method based on CTCAE hierarchical driving further includes: The element value table, evidence location identifier, degree element value, severity classification result, corresponding terminology, ordinal scale value, degree of difference, causal relationship arbitration result, and the bound configuration version, causal relationship determination result or pending mark of each independent algorithm path are written to the audit log in an append-only manner. Records already written to the audit log must not be overwritten. When the configuration version changes, steps S1 to S4 are re-executed for the processed adverse event candidate events bound to the configuration version involved in the change, based on the changed configuration version, to obtain the recalculation results. An impact list is generated that records the changes in the conclusions. The impact list includes adverse event candidate events, the severity classification results or causal relationship arbitration results before and after the change, and the reasons for the change. The recalculation results and the impact list are written to the audit log in an append-only manner. When it is necessary to retain two conclusions before and after the change, a new event version record is created and the old record is frozen. The record is marked as a cross-version related record in the audit log.

[0061] Understandably, the audit log records all intermediate and final results generated during the processing of steps S1 to S4, covering the complete chain from term matching, element value selection, degree element synthesis, severity classification, to the judgment results of the three independent algorithm paths, ordinal scale mapping, difference degree calculation, and finally to the causal association arbitration result. It also clearly specifies the configuration version on which it is based, ensuring that the processing conclusion of any candidate event can be traced back to the specific judgment basis and configuration version. The writing method is append-only, meaning new records can only be appended, and previously written historical records cannot be modified or deleted, thus ensuring that past conclusions are not overwritten or tampered with by subsequent operations. When the configuration versions of CTCAE grading tables, terminology mapping tables, etc., change, this invention does not directly overwrite the conclusions already obtained under the old version with the new version. Instead, it re-executes the complete steps S1 to S4 for the processed candidate events based on the changed configuration version to obtain recalculated results. The recalculated results and the differences between the conclusions before and after the change are then used to generate an impact list, which is written to the audit log, enabling the processing conclusions to be comparable and traceable before and after the configuration version change.

[0062] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-path causal association arbitration method based on CTCAE hierarchical driving, characterized in that, include: S1. Obtain candidate adverse events and their evidence texts, match the candidate adverse events with candidate terms in the CTCAE grading table, determine the corresponding terms, and retrieve the common evaluation element table and grading judgment rules associated with the corresponding terms. S2. Extract the values ​​of the observed elements from the common evaluation element table from the evidence text and verify them. Set the values ​​that fail the verification to unknown to obtain the element value table. The observed element values ​​in the element value table are synthesized according to the synthesis rules recorded in the common evaluation element table to obtain the degree element values; the element value table and degree element values ​​are substituted into the judgment expression of the grading judgment rule to calculate the value, and each severity level is marked according to whether it has been satisfied, excluded, or not excluded to determine the severity grading result. S3. Input the element value table into multiple independent algorithm paths that do not share intermediate conclusions and perform parallel computation. Each algorithm path corresponds to a different judgment rule system. At least one algorithm path uses a weighted summation of the observed element values ​​to determine the result, and at least one algorithm path uses a conditional judgment method for each observed element value to determine the result. When the values ​​of the observed elements involved in the judgment are insufficient to uniquely determine the result, the corresponding algorithm path outputs an undetermined mark; otherwise, it outputs the causal association judgment result. S4. Determine the causal association arbitration result based on the causal association determination results or pending markers output by each algorithm path: When there are pending markers in the output of each algorithm path, the causal association arbitration result will be set to be transferred to manual review; When there are no pending labels, the causal association judgment results output by each algorithm path are mapped to a unified ordinal scale and grouped and classified. The degree of difference is determined according to whether the classification results are the same. When the degree of difference is zero, it is set to automatic adoption. When the degree of difference is greater than zero and does not exceed the preset threshold, it is set to low confidence adoption. When the degree of difference exceeds the preset threshold, it is set to manual review. S5. Output the severity grading results and the causal relationship arbitration results.

2. The multi-path causal association arbitration method based on CTCAE hierarchical driving as described in claim 1, characterized in that: Step S1 specifically includes: Read the pre-bound configuration version of the adverse event candidate event. The configuration version includes at least the CTCAE classification table version, term mapping table version, thesaurus version, sentence vector model identifier and similarity threshold. Determine the candidate term set in the published CTCAE classification table corresponding to the configuration version. The names of adverse event candidate events are compared with the names of candidate terms in the candidate term set to obtain exact match terms; when there are no exact match terms, the names of adverse event candidate events are compared with the synonyms of candidate terms in the candidate term set to obtain synonym extended match terms. When both exact match terms and synonym expansion terms are empty, semantic similarity is calculated based on the definition and description of the candidate terms in the evidence text and the candidate term set to obtain the semantic similarity of each candidate term; when the maximum value of semantic similarity is lower than the similarity threshold of the configured version, the adverse event candidate event is marked as term cannot be located and the automatic classification ends. When a candidate term exists that is not empty in an exact match, or is not empty in a synonym expansion match, or has a maximum semantic similarity of not less than the similarity threshold, the candidate term is identified as the corresponding term. The common evaluation element table associated with the corresponding term in the published CTCAE grading table and the grading determination rules are retrieved. The common evaluation element table includes observation elements, degree elements and the synthesis rules of degree elements. The grading determination rules record the determination expressions for each severity level.

3. The multi-path causal association arbitration method based on CTCAE hierarchical driving as described in claim 2, characterized in that: Step S2 specifically includes: S21. Determine the description judgment results and evidence location markers of each observation element from the evidence text according to the preset set of acceptable description forms. S22. Verify the description judgment result and evidence location identifier with the evidence text. If the verification fails, set the value of the observed element to unknown. If the verification passes, write the value indicated by the description judgment result into the observed element to obtain the element value table. S23. According to the synthesis rules recorded in the common evaluation element table, the degree element value is obtained from the observed element value in the element value table; S24. Substitute the element value table and the degree element value into the judgment expression recorded in the grading judgment rule to obtain the severity grading result.

4. The multi-path causal association arbitration method based on CTCAE hierarchical driving as described in claim 3, characterized in that: Step S22 specifically includes: Read the text fragment pointed to by the evidence location marker, determine whether the text fragment falls within the scope of the selected description form or the registered synonyms of the selected description form in the description judgment result, and obtain the location consistency judgment result; If the location consistency determination result is inconsistent, or the description determination result does not belong to the preset set of acceptable description forms, or the content indicated by the description determination result is inconsistent with the text fragment pointed to by the evidence location marker, the verification is determined to be unsuccessful and the value of the observed element is set to unknown; otherwise, the verification is determined to be successful and the value indicated by the description determination result is written into the element value table. For numerical observation features, the verification method is determined based on the type of data source: For laboratory numerical observation elements derived from structured test reports, the test report values ​​are directly written into the element value table and the source location of the values ​​is retained. For laboratory numerical observation elements extracted from the evidence text, an evidence location identifier is given, and the description judgment result and the evidence location identifier are verified with the evidence text. If the verification fails, the laboratory numerical value is set to unknown.

5. The multi-path causal association arbitration method based on CTCAE hierarchical driving as described in claim 4, characterized in that: Step S24 specifically includes: Substitute the numerical observation elements and degree elements from the element value table into the judgment expression recorded in the grading judgment rule, and mark them as satisfied, excluded, or not yet excluded due to unknown values ​​according to each severity level. The highest-ranking severity level that is marked as satisfied is... The highest-ranking severity level among those marked as not yet ruled out is... ,in This indicates the highest severity level of the status that has been met. This indicates the highest severity level among the states that have not yet been ruled out; when it is higher than... When all possible severity levels have been excluded, the severity classification result will be taken as a point level. When higher than If any severity level has not yet been ruled out, the severity classification result will be taken as an interval. When there are no satisfied levels but there are levels that have not yet been excluded, the severity classification result is represented by an upper bound of ; The interval with an undetermined lower bound.

6. The multi-path causal association arbitration method based on CTCAE hierarchical driving as described in claim 5, characterized in that: The multiple algorithm paths are specifically a first independent algorithm path, a second independent algorithm path, and a third independent algorithm path; step S3 specifically includes: S31. Assign scores to the observed element values ​​in the element value table according to the preset weighted scoring rules and sum them to obtain the weighted total score S. Determine the causal relationship judgment result or pending label of the first independent algorithm path based on the preset score range of S. S32. According to the preset multi-dimensional condition judgment rules, judge the values ​​of the observed elements in the element value table one by one, and determine the causal relationship judgment result or pending mark of the second independent algorithm path. S33. Match the values ​​of observed elements in the element value table according to the rule chain formed by the correspondence between conditions and conclusions. The correspondence between conditions and conclusions is compiled from the clinical trial protocol. Determine the causal association judgment result or pending label of the third independent algorithm path according to the triggered correspondence between conditions and conclusions.

7. The multi-path causal association arbitration method based on CTCAE hierarchical driving as described in claim 6, characterized in that: The parallel operations specifically include: When there are observation elements with unknown values ​​in the first independent algorithm path, for each observation element with an unknown value, the values ​​included in the preset allowed value range are substituted one by one into the weighted scoring rule to calculate the total score, resulting in multiple total scores; the minimum total score among the multiple total scores is recorded as . The maximum total score is recorded as To obtain the total score range When the total score interval crosses the correlation level boundary corresponding to the preset score interval, the output of the first independent algorithm path is set to the pending mark; when the total score interval falls within the same correlation level, the output of the first independent algorithm path is set to the corresponding causal correlation category. When the value of the observed element participating in the judgment in the second independent algorithm path is unknown, or the judgment path falls into the preset multi-level condition branches and is marked as a branch that cannot be classified or evaluated, the output of the second independent algorithm path is set to the pending label; otherwise, it is set to the causal association category corresponding to the judgment path. When the observation element required by the rule chain in the third independent algorithm path is missing or has an unknown value in the element value table, or when the rule chain determines that the data is insufficient, the output of the third independent algorithm path is set to the pending label; otherwise, it is set to the causal association category triggered by the rule chain.

8. The multi-path causal association arbitration method based on CTCAE hierarchical driving as described in claim 7, characterized in that: The step of mapping the causal association determination results output by each algorithm path to a unified ordinal scale is as follows: mapping is performed according to the causal association determination results and ordinal scale value lookup table bound to the configuration version, and the mapped ordinal scale values ​​are grouped and classified according to the preset association level. The configuration version is the version identifier pre-bound to the adverse event candidate event, and the lookup table remains unchanged after the configuration version is released. When the association level groups to which each algorithm path belongs are the same, the degree of difference is recorded as zero; when the association level groups to which each algorithm path belongs are different, the difference between the maximum and minimum values ​​of the ordinal scale values ​​output by each algorithm path is taken as the degree of difference.

9. The multi-path causal association arbitration method based on CTCAE hierarchical driving as described in claim 1, characterized in that: Step S1 is preceded by: A consistency check is performed on the degree elements and the judgment expressions in the grading judgment rules recorded in the common evaluation element table. The following situations are marked as defects, and the consistency check results are obtained: degree elements that hit the preset degree vocabulary but are not included in the exemption list; judgment expressions that cannot be satisfied; degree elements that have been defined but have never been referenced by any judgment expression. When the consistency verification result does not contain any defects, the shared evaluation element table, the grading judgment rules, the preset set of acceptable description forms, and the list of exemption items will be published as the published version; if it contains defects, it will not be published. To bind a configuration version to an adverse event candidate, if there is no corresponding published version for the configuration version, the adverse event candidate must not be matched with the CTCAE classification table.

10. The multi-path causal association arbitration method based on CTCAE hierarchical driving as described in claim 1, characterized in that: Also includes: The element value table, evidence location identifier, degree element value, severity classification result, corresponding terminology, ordinal scale value, degree of difference, causal relationship arbitration result, and the bound configuration version, causal relationship determination result or pending mark of each independent algorithm path are written to the audit log in an append-only manner. Records already written to the audit log must not be overwritten. When the configuration version changes, steps S1 to S4 are re-executed for the processed adverse event candidate events bound to the configuration version involved in the change, based on the changed configuration version, to obtain the recalculation result and generate an impact list recording the change in conclusion. The impact list includes adverse event candidate events, severity classification results or causal relationship arbitration results before and after the change, and the reason for the change. The recalculation result and the impact list are written to the audit log in an append-only manner. When it is necessary to retain both the original and revised versions of the conclusions, create a new event version record and freeze the old record, marking it as a cross-version related record in the audit log.

Images