A method and system for automatically generating electronic nursing records

Abstract

This invention discloses a method and system for automatically generating electronic nursing records, belonging to the technical field of automatic generation. Specifically, it includes segmenting and assigning unique evidence identifiers to the unified raw data to obtain evidence fragments; encapsulating all evidence fragments into a closed-set evidence set; generating a draft of the electronic nursing record carrying evidence identifiers and splitting it into declaration units; performing bidirectional consistency verification between the closed-set evidence set and the declaration units; handling mandatory evidence coverage anomalies and implementing a missing evidence compensation generation mechanism to obtain a compensation processing draft; based on preset type rules, distributing the declaration units to be verified in the compensation processing draft to different verification routes for verification, obtaining a ternary judgment result and recording the judgment log entries; updating the declaration units according to the ternary judgment result to obtain the final draft of the electronic nursing record and performing a final draft balancing review; and outputting the manually intervened final draft of the electronic nursing record after updating with manual intervention instructions.

Description

Technical Field

[0001] This invention relates to the field of automatic generation technology, and in particular to a method and system for automatically generating electronic nursing records. Background Technology

[0002] Electronic nursing records are indispensable medical documents in clinical nursing. They objectively, truthfully, and continuously reflect changes in a patient's condition, vital signs, nursing assessment results, and the implementation of various nursing interventions. While large language models have been widely used in recent years to assist in generating medical records and reduce the workload of medical staff, they have revealed three fatal flaws in demanding clinical scenarios: First, the risk of fabrication without prior knowledge—models are susceptible to parameter memory issues, fabricating vital sign values ​​or nursing interventions that did not occur; second, the risk of omitting key medical conditions—existing one-way fact-checking only reviews generated text and cannot detect high-risk abnormal data missed by the model due to induction, which can easily lead to medical accidents; and third, the risk of excessive subjective inference—models often exaggerate objective behavior into subjective conclusions. Therefore, existing technologies cannot simultaneously achieve absolute traceability, zero omissions, and objectivity and rigor, necessitating a new automatic generation method with two-way closed-loop constraints and the ability to prevent excessive inference.

[0003] Currently, Chinese invention patent CN114242196B discloses a method and apparatus for automatically generating clinical medical records. The method includes: acquiring electronic medical records and corresponding time information; segmenting the electronic medical records using a word segmentation library based on diagnostic information to obtain case keywords; constructing spatiotemporal features using each electronic medical record as a node, case keywords as the spatial information of the node, and the time interval of the electronic medical records as the time information; training a spatiotemporal sequence model using n consecutive sequence data from the spatiotemporal features as training samples and case keywords as combined labels; outputting predicted case keywords for the predicted time object based on the trained spatiotemporal sequence model and the time interval of the predicted time; comparing the predicted keywords with the detection report detected at the predicted time to generate clinical medical text. However, this technology struggles to prevent unsubstantiated statements and misquoted citations in nursing record generation, lacking traceability, verifiability, and completeness of the generated content. Summary of the Invention

[0004] The technical problem solved by this invention is that related technologies are unable to prevent the generation of nursing records from containing unsubstantiated statements and misquoted references, and lack the traceability, verifiability and completeness of the generated content.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] A method for automatically generating electronic nursing records includes the following steps:

[0007] Step S1: The raw data after unification is segmented and assigned a unique evidence identifier to obtain evidence fragments, and all evidence fragments are encapsulated into a closed set of evidence.

[0008] Step S2: Generate a draft of the nursing electronic record carrying evidence identifier based on the closed set of evidence, and break it down into declaration units;

[0009] The closed set of evidence and the declaration unit are subjected to bidirectional consistency verification. An exception of mandatory evidence not being covered is handled and a mechanism for generating missing evidence compensation is implemented. The compensation processing draft is then reorganized.

[0010] Step S3: Based on the preset type rules, the declaration units to be verified in the compensation processing draft are distributed to different verification routes for verification, and the ternary judgment results are obtained and the judgment log entries are recorded.

[0011] Step S4: Update the declaration unit according to the ternary judgment result to obtain the final draft of the nursing electronic record and perform final draft balancing and verification. After updating with the manual intervention instruction, output the final draft of the nursing electronic record after manual intervention.

[0012] Preferably, step S1 specifically includes:

[0013] Step S11: Obtain the raw data and perform unified processing on the raw data. The unified processing includes time alignment, unit alignment, and preservation of text transcription results.

[0014] The raw data includes monitoring data, medical orders, voice interaction data, and manually entered data;

[0015] The monitoring data includes heart rate, blood pressure, and blood oxygen saturation;

[0016] Step S12: Divide the unified raw data into evidence fragments, assign a unique evidence identifier to each evidence fragment, and establish a bidirectional index table between the evidence identifier and the content of the evidence fragment through a key-value pair mapping table.

[0017] Step S13: Evaluate each piece of evidence using preset risk rules;

[0018] Set the mandatory carry-over label for evidence fragments that meet the preset risk rules to 1;

[0019] Set the mandatory carry-over label of evidence fragments that do not comply with the preset risk rules to 0;

[0020] Step S14: Encapsulate all evidence fragments into a closed set of evidence, wherein the evidence fragments include evidence identifiers and mandatory carry-over label fields.

[0021] Preferably, step S2 specifically includes:

[0022] Step S21: Retrieve the evidence fragment content corresponding to the evidence identifier in the closed set of evidence through the bidirectional index table, and perform anchor point constraint operation on the closed set of evidence based on the evidence fragment content to obtain the nursing electronic record draft, wherein the nursing electronic record draft carries the evidence identifier.

[0023] Step S22: Divide the draft of the nursing electronic record into declaration units according to the sentence segmentation, take the text content in the draft of the nursing electronic record as the text content of the declaration unit, and extract the evidence identifier set of each declaration unit in the draft of the nursing electronic record.

[0024] Step S23: Perform a two-way consistency check between the closed set of evidence and the declaration unit to obtain a draft of the compensation process;

[0025] Step S24: Using a bidirectional index table, retrieve the evidence fragment content corresponding to the evidence identifier set carried by the statement unit to be tested in the compensation processing draft, and obtain the evidence fragment content set corresponding to the statement unit to be tested.

[0026] Preferably, the bidirectional consistency check specifically includes:

[0027] Obtain the set of evidence identifiers for the claim units of the draft nursing electronic records, and perform a claim unit consistency check:

[0028] When it is detected that the evidence identifier set of the declaration unit is empty, or that any evidence identifier in the evidence identifier set does not belong to the closed set of evidence, the declaration unit is deleted;

[0029] Otherwise, retain the declaration unit;

[0030] Get the collection of declaration units after deletion;

[0031] By summing the evidence identifier sets of the declared units in the deleted set of declared units, we obtain the covered evidence identifier set;

[0032] Extract the evidence identifier with the mandatory carry-over label 1 from the closed set of evidence as the target evidence identifier, and then perform a target evidence identifier retrieval in the covering evidence identifier set:

[0033] When the target evidence identifier is found to be included in the set of covered evidence identifiers, the target evidence identifier is marked as successfully retrieved.

[0034] When the target evidence identifier is not found in the covered evidence identifier set, the target evidence identifier is determined to be an abnormal forced evidence not covered, recorded in the uncovered target evidence identifier set, and the missing evidence compensation generation mechanism is executed.

[0035] When the search is completed and the target evidence identifier set is empty, the claim units in the deleted claim unit set are reorganized in the order of the nursing electronic record draft to obtain the compensation processing draft.

[0036] Preferably, the mechanism for generating compensation for missing evidence includes:

[0037] The evidence fragment content corresponding to the target evidence identifier is retrieved through a bidirectional index table. The supplementary declaration unit is generated using the target evidence identifier and the corresponding evidence fragment content as input information sources.

[0038] Obtain the set of evidence identifiers for the supplementary declaration unit as the supplementary evidence identifier set, and check whether the supplementary evidence identifier set includes the target evidence identifier input when generating the supplementary declaration unit;

[0039] If the check result indicates that the supplementary declaration unit is not included, delete the unit.

[0040] When the inspection result indicates inclusion, a supplementary statement unit will be added to the draft of the electronic nursing record;

[0041] The bidirectional consistency check was re-performed on the draft of the nursing electronic record after the addition of supplementary declaration units to obtain the updated set of uncovered target evidence identifiers.

[0042] like If so, the mechanism for generating compensation for missing evidence will end;

[0043] Otherwise, if Then the mechanism for generating compensation for missing evidence will end, and will Add to the list of tasks requiring manual intervention;

[0044] otherwise, ,Will Defined as Repeated execution of the missing evidence compensation generation mechanism;

[0045] in, The set of uncovered target evidence identifiers obtained after the k-th round of bidirectional consistency verification is obtained by re-performing the bidirectional consistency verification after the k-th round of compensation generation and the addition of supplementary declaration units. , The cardinality of the uncovered target evidence identifier set, It is an empty set;

[0046] The declaration units in the nursing electronic record draft at the end of the missing evidence compensation generation mechanism are reorganized in the order of the nursing electronic record draft to obtain the compensation processing draft.

[0047] Preferably, step S3 specifically includes:

[0048] Step S31: According to the preset type rules, generate a declaration type tag for the declaration unit to be checked in the compensation processing draft. The declaration type includes determining the declaration type and judging the declaration type.

[0049] Step S32: Generate routing results based on the declaration type marker for the declaration unit to be verified. The routing results include rule verification routes and adversarial verification routes, specifically including:

[0050] When the declaration type is marked as a definitive declaration type, the generated route result is a rule-verified route;

[0051] When the declaration type is marked as a judgment declaration type, the generated route result is a route that resists verification;

[0052] Step S33: When the routing result is a rule-verified route, the rule verification is performed on the declaration unit to be verified to obtain a ternary judgment result. The ternary judgment result includes only supporting declaration units, only supporting counterfactual declaration units, and insufficient evidence.

[0053] The rule verification processing logic includes:

[0054] Extract verifiable fields from the declaration unit to be inspected, extract evidence fields of the same type as the verifiable fields from the evidence fragment content set, and perform field consistency verification according to the preset verification rules;

[0055] When the verifiable field matches the evidence field, the output ternary judgment result is "only supports declaration unit".

[0056] When the evidence field exists but is inconsistent with the verifiable field, the output ternary judgment result is that only the counterfactual declaration unit is supported.

[0057] When it is impossible to extract evidence fields of the same type as the verifiable fields, the output ternary judgment result is insufficient evidence;

[0058] When the routing result is a route that resists verification, the unit to be verified is subjected to resist verification to obtain a ternary determination result;

[0059] By combining the text content of the statement unit to be tested with the statement type tag, a counterfactual statement unit is generated, and a triplet is constructed. The triplet includes a set of evidence fragments, the statement unit to be tested, and the counterfactual statement unit.

[0060] Given a set of evidence fragments, the evidence support judgment is performed on the unit of statement to be examined and the counterfactual unit, respectively. The processing expression is as follows:

[0061] ;

[0062] ;

[0063] in, To determine whether the set of evidence fragments supports the declaration unit. As a result, For the unit of declaration to be tested, As a counterfactual statement unit, , For supportive decision values, , 1 indicates support, 0 indicates no support;

[0064] The ternary decision result is output based on the support decision value, specifically including:

[0065] when =1 and When the value is 0, the output ternary determination result is "only supports declaration units";

[0066] when =0 and When the value is 1, the output ternary determination result is that only counterfactual declaration units are supported;

[0067] Otherwise, the ternary determination result is insufficient evidence;

[0068] Step S34: Record the declaration unit identifier, the declaration type tag of the declaration unit, the routing result, the evidence identifier set, the evidence fragment content set, and the ternary judgment result as judgment log entries, and bind the judgment log entries to the declaration unit identifier.

[0069] Preferably, step S4 specifically includes:

[0070] Step S41: Perform update operations on the declaration units to be tested in the compensation processing draft according to the ternary determination results. The update operations include retention, deletion and downgrade rewriting.

[0071] Step S42: When the ternary determination result is that only the declaration unit is supported, the declaration unit to be tested is retained unchanged;

[0072] When the ternary determination result indicates that only counterfactual statement units are supported, the statement unit to be tested is deleted and written to the determination log entry;

[0073] When the ternary triad result indicates insufficient evidence, evidence identification is corrected to obtain an updated ternary triad result. An update operation is then performed based on the updated ternary triad result, specifically including:

[0074] When the updated ternary determination result is that only the declaration unit is supported, the declaration unit is retained and its evidence identifier set is updated to the corrected evidence identifier set.

[0075] When the updated ternary determination result is not "only supports declaration units", the declaration unit is downgraded and rewritten to generate an alternative declaration unit.

[0076] When the replacement declaration unit is successfully generated, the declaration unit is replaced with the replacement declaration unit.

[0077] If the generation of a replacement declaration unit fails, the declaration unit is deleted.

[0078] The generation of alternative declaration units specifically includes:

[0079] The set of evidence fragments to be downgraded and rewritten is retrieved from the set of evidence identifiers of the declaration unit to be downgraded and rewritten by the bidirectional index table. The set of evidence fragments to be downgraded and rewritten is sorted according to the time order of the evidence fragments. The sorted evidence fragments to be downgraded and rewritten are then concatenated to generate the alternative declaration unit.

[0080] Step S43: Reorganize all the declaration units processed in S42 according to the order of the nursing electronic record draft to obtain the final version of the nursing electronic record;

[0081] Step S44: Perform a final balance check on the final draft of the electronic nursing record;

[0082] Extract the target evidence identifier with the mandatory carry-over label 1 from the closed set of evidence; retrieve the evidence identifier set of the statement unit to be examined based on the final draft of the nursing electronic record.

[0083] If at least one set of evidence identifiers includes the target evidence identifier, then the target evidence identifier is marked as covered.

[0084] If no evidence identifier set contains the target evidence identifier, the retrieval is deemed a failure, and the target evidence identifier is added to the manual intervention pending list.

[0085] Step S45: Output the list of pending manual interventions, the final draft of the electronic nursing record, and the judgment log entries;

[0086] Step S46: Receive manual intervention instructions, process the target declaration unit based on the manual intervention instructions, and output the final draft of the nursing electronic record after manual intervention;

[0087] The manual intervention instructions include restoration, replacement, and maintenance operations for the target declaration unit.

[0088] Preferably, the correction of evidence identification specifically includes:

[0089] Extract the text semantic features of the statement unit to be tested, and traverse the text semantic features of each evidence fragment in the closed set of evidence. Calculate the semantic relevance score, arrange the evidence identifiers in descending order of semantic relevance score, and extract the first N evidence identifiers to form a candidate evidence identifier set, where N is a positive integer.

[0090] Select at least one candidate evidence identifier from the candidate evidence identifier set to form a modified evidence identifier set, construct the modified evidence identifier set, update the evidence fragment content set according to the modified evidence identifier set, and obtain the modified evidence fragment content set;

[0091] The corrected set of evidence identifiers is used to replace the set of evidence identifiers, and the corrected set of evidence fragment content is used to replace the set of evidence fragment content. Corresponding route verification is then performed to obtain the updated ternary judgment result.

[0092] Preferably, step S45 specifically includes:

[0093] If the manual intervention instruction is a recovery operation, the deleted and replaced declaration units corresponding to the target declaration unit will be written into the final version of the nursing electronic record and used as the declaration unit after the manual intervention.

[0094] If the manual intervention instruction is a replacement operation, then the corresponding declaration unit in the final draft of the nursing electronic record will be replaced with the replacement text provided by the manual intervention instruction;

[0095] If the manual intervention instruction is to maintain the operation, then the final version of the electronic nursing record remains unchanged.

[0096] An automatic generation system for nursing electronic records includes a data acquisition module, a verification module, a processing module, and an output module;

[0097] The acquisition module is used to segment the unified raw data and assign a unique evidence identifier to obtain evidence fragments, and encapsulate all evidence fragments into a closed set of evidence.

[0098] The verification module is used to generate a draft of the nursing electronic record carrying evidence identifier based on the closed set of evidence, and to break it down into declaration units;

[0099] The closed set of evidence and the declaration unit are subjected to bidirectional consistency verification. An exception of mandatory evidence not being covered is handled and a mechanism for generating missing evidence compensation is implemented. The compensation processing draft is then reorganized.

[0100] The processing module is used to preset routing rules, distribute the declaration units to be verified in the compensation processing draft to different verification routes for verification, obtain a ternary judgment result and record the judgment log entries.

[0101] The output module is used to update the declaration unit according to the ternary judgment result, obtain the final draft of the nursing electronic record, perform final draft balancing and verification, and output the final draft of the nursing electronic record after manual intervention after updating with the manual intervention instruction.

[0102] The beneficial effects of this invention are as follows: This invention establishes the generation of nursing electronic records based on evidence identification constraints and two-way consistency verification, and automatically triggers compensation generation when uncovered key evidence is found until the coverage conditions are met or manual intervention is initiated. The generated content is linked one-to-one with evidence fragments, facilitating traceability and auditing; declaration units with missing evidence identification or illegal citations are deleted, reducing the risk of unprovoked fabrication; key evidence coverage is retrieved based on mandatory carry-over tags, and the uncovered set is output, reducing omissions of high-risk events; compensation generation and duplicate verification form a closed loop, improving record integrity and providing a terminateable abnormal exit mechanism; a three-element judgment result is output based on type routing and counterfactual verification, reducing inference expansion and overly strong wording; combined with final draft review and manual instruction updates, the final draft is controllable and records of manual decisions are preserved. Attached Figure Description

[0103] Figure 1 This is a basic flowchart illustrating a method for automatically generating electronic nursing records according to an embodiment of the present invention.

[0104] Figure 2 This is a basic flowchart of an automatic nursing electronic record generation system provided in one embodiment of the present invention;

[0105] Figure 3 This is a flowchart of the consistency detection process for declared units in bidirectional consistency verification.

[0106] Figure 4 Flowchart of the back-end of bidirectional consistency verification Detailed Implementation

[0107] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0108] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for automatically generating electronic nursing records is provided, comprising the following steps:

[0109] Step S1: The raw data after unification is segmented and assigned a unique evidence identifier to obtain evidence fragments, and all evidence fragments are encapsulated into a closed set of evidence.

[0110] Step S2: Generate a draft of the nursing electronic record carrying evidence identifier based on the closed set of evidence, and break it down into declaration units;

[0111] The closed set of evidence and the declaration unit are subjected to bidirectional consistency verification. An exception of mandatory evidence not being covered is handled and a mechanism for generating missing evidence compensation is implemented. The compensation processing draft is then reorganized.

[0112] Step S3: Based on the preset routing rules, the declaration units to be verified in the compensation processing draft are distributed to different verification routes for verification, and the three-element judgment result is obtained and the judgment log entries are recorded.

[0113] Step S4: Update the declaration unit according to the ternary judgment result to obtain the final draft of the nursing electronic record and perform final draft balancing and verification. After updating with the manual intervention instruction, output the final draft of the nursing electronic record after manual intervention.

[0114] This invention establishes the generation of electronic nursing records based on evidence identification constraints and bidirectional consistency verification. It automatically triggers compensatory generation upon discovering uncovered critical evidence, continuing until coverage conditions are met or manual intervention is initiated. The generated content is linked one-to-one with evidence fragments, facilitating traceability and auditing. Declarations with missing evidence identification or illegal citations are deleted, reducing the risk of unprovoked fabrication. The mandatory carry-over tag system retrieves the coverage of critical evidence and outputs the uncovered set, reducing omissions of high-risk events. Compensatory generation and duplicate verification form a closed loop, improving record integrity and providing a terminateable exit mechanism. Based on type routing and counterfactual verification, a ternary judgment result is output, reducing inferences and overly strong wording. Combined with final draft review and manual instruction updates, the final draft is controllable, and records of manual decisions are preserved.

[0115] In a specific embodiment, step S1 specifically includes:

[0116] Step S11: Obtain the raw data and perform unified processing on the raw data. The unified processing includes time alignment, unit alignment, and preservation of text transcription results.

[0117] Raw data includes monitoring data, medical orders, voice interaction data, and manually entered data;

[0118] The monitoring data includes heart rate, blood pressure, and blood oxygen saturation;

[0119] Step S12: Divide the unified raw data into evidence fragments, assign a unique evidence identifier to each evidence fragment, and establish a bidirectional index table between the evidence identifier and the content of the evidence fragment through a key-value pair mapping table.

[0120] Step S13: Evaluate each piece of evidence using preset risk rules;

[0121] Set the mandatory carry-over label for evidence fragments that meet the preset risk rules to 1;

[0122] Set the mandatory carry-over label of evidence fragments that do not comply with the preset risk rules to 0;

[0123] Step S14: Encapsulate all evidence fragments into a closed set of evidence, which includes evidence identifier and mandatory carry-over label fields.

[0124] Specifically, the standardization process includes time alignment, unit alignment, and preservation of text transcription results. Time alignment establishes a unified time base for the original data and transforms the time representation of various original data records to a time representation under this unified time base. Preferably, data records lacking timestamps are associated with the time range of their collection period, and the corresponding time information field is stored in the data body.

[0125] Unit alignment involves identifying unit information associated with numerical fields in the original data and converting the numerical fields to standard units before storage. Numerical fields include values ​​related to vital signs, intake and output, drug dosage, and laboratory indicators; ideally, both the original and converted unit information should be retained during the conversion process.

[0126] When the original data includes voice interaction data, the voice interaction data is processed to generate a text transcription result. The text transcription result is retained as part of the original data, and the text transcription result is stored in association with its corresponding time information, bed information, and patient identification information.

[0127] The construction of a bidirectional index table for evidence identifiers and evidence fragment content by building a key-value pair mapping table specifically includes: assigning a unique evidence identifier to each evidence fragment and establishing a mapping relationship from evidence identifier to evidence fragment content in the mapping table, enabling the retrieval of the corresponding evidence fragment content through the evidence identifier; establishing an index relationship from evidence fragment content to evidence identifier in the mapping table, enabling retrieval based on evidence fragment content and returning the corresponding set of evidence identifiers; and storing the bidirectional index table in association with the closed set of evidence for subsequent use in generating draft nursing electronic records, performing consistency checks, correcting evidence identifiers, and retrieving the set of evidence fragment content.

[0128] The default risk rule is that when evaluating each piece of evidence, if the content of the evidence piece contains the value of the corresponding monitoring field and meets the risk threshold condition, the mandatory carry-over label of the evidence piece is set to 1; if the evidence piece does not contain the monitoring field corresponding to the risk threshold condition or the content of the evidence piece does not meet the risk threshold condition, the system sets the mandatory carry-over label of the evidence piece to 0.

[0129] The monitored fields include heart rate, blood pressure, and blood oxygen saturation;

[0130] Risk threshold conditions include blood oxygen saturation below 90%, heart rate above 120 beats / min, heart rate below 50 beats / min, and systolic blood pressure below 90 mmHg.

[0131] In a specific embodiment, step S2 specifically includes:

[0132] Step S21: Retrieve the evidence fragment content corresponding to the evidence identifier in the closed set of evidence through the bidirectional index table, and perform anchor point constraint operation on the closed set of evidence based on the evidence fragment content to obtain the nursing electronic record draft, wherein the nursing electronic record draft carries the evidence identifier.

[0133] Step S22: Divide the draft of the nursing electronic record into declaration units according to the sentence segmentation, take the text content in the draft of the nursing electronic record as the text content of the declaration unit, and extract the evidence identifier set of each declaration unit in the draft of the nursing electronic record.

[0134] Step S23: Perform a two-way consistency check between the closed set of evidence and the declaration unit to obtain a draft of the compensation process;

[0135] Step S24: Using a bidirectional index table, retrieve the evidence fragment content corresponding to the evidence identifier set carried by the statement unit to be tested in the compensation processing draft, and obtain the evidence fragment content set corresponding to the statement unit to be tested.

[0136] Specifically, the anchor point constraint operation on the closed set of evidence based on the content of evidence fragments includes:

[0137] Using a bidirectional index table, the system retrieves the evidence fragments corresponding to each evidence identifier in the closed-set evidence collection. Based on these evidence fragments, it generates a draft of the electronic nursing record and applies anchor constraints to the generation process. These constraints include associating each clinical statement in the draft with an evidence identifier and explicitly embedding the evidence identifier into the draft, ensuring that the draft carries the evidence identifier. The system then performs a citation integrity check on the draft, which checks whether each clinical statement carries an evidence identifier. When the system detects a clinical statement without an evidence identifier, it deletes the statement or marks it as unusable.

[0138] Nursing electronic record drafts typically consist of multiple relatively independent statements. Sentence segmentation can divide the nursing electronic record draft into the smallest statement granularity that facilitates line-by-line verification and updating without introducing additional semantic inference. It is preferable to use punctuation marks and line breaks as the basis for sentence segmentation, splitting the nursing electronic record draft into multiple statement units, and extracting evidence identifier sets, performing consistency checks, and performing subsequent verification and update operations for each statement unit.

[0139] To avoid losing semantic context after segmentation, the system prefers to retain the original order information of the declaration units in the nursing electronic record draft when splitting the declaration units, and saves the text fragment corresponding to each declaration unit as the text content of that declaration unit.

[0140] The compensation processing draft is the draft form of the nursing electronic record obtained after the system performs bidirectional consistency verification and missing evidence compensation processing. To prevent confusion in the use of qualifiers for declaration units in subsequent steps, each declaration unit in the compensation processing draft is defined as a declaration unit to be tested.

[0141] In a specific embodiment, bidirectional consistency verification includes:

[0142] Obtain the set of evidence identifiers for the claim units of the draft nursing electronic records, and perform a claim unit consistency check:

[0143] When it is detected that the evidence identifier set of the declaration unit is empty, or that any evidence identifier in the evidence identifier set does not belong to the closed set of evidence, the declaration unit is deleted;

[0144] Otherwise, retain the declaration unit;

[0145] Get the collection of declaration units after deletion;

[0146] By summing the evidence identifier sets of the declared units in the deleted set of declared units, we obtain the covered evidence identifier set;

[0147] Extract the evidence identifier with the mandatory carry-over label 1 from the closed set of evidence as the target evidence identifier, and then perform a target evidence identifier retrieval in the covering evidence identifier set:

[0148] When the target evidence identifier is found to be included in the set of covered evidence identifiers, the target evidence identifier is marked as successfully retrieved.

[0149] When the target evidence identifier is not found in the covered evidence identifier set, the target evidence identifier is determined to be an abnormal forced evidence not covered, recorded in the uncovered target evidence identifier set, and the missing evidence compensation generation mechanism is executed.

[0150] When the search is completed and the target evidence identifier set is empty, the claim units in the deleted claim unit set are reorganized in the order of the nursing electronic record draft to obtain the compensation processing draft.

[0151] Specifically, refer to Figure 3 and Figure 4 This is the overall flowchart for bidirectional consistency verification;

[0152] The uncovered target evidence identifier set is a dynamically generated set during the target evidence identifier retrieval process. First, evidence identifiers with a mandatory carry-over tag of 1 are extracted from the closed-set evidence set to form the target evidence identifier set. Then, the evidence identifier sets carried by all claim units in the current nursing electronic record draft are aggregated to form the covered evidence identifier set. The uncovered target evidence identifier set is initialized as an empty set, and a retrieval is performed for each target evidence identifier in the target evidence identifier set: if the covered evidence identifier set contains the target evidence identifier, the target evidence identifier is marked as successfully retrieved; if the covered evidence identifier set does not contain the target evidence identifier, the target evidence identifier is added to the uncovered target evidence identifier set. After completing the retrieval of all target evidence identifiers, if the uncovered target evidence identifier set is not empty, it is used as the input for the missing evidence compensation generation mechanism; if the uncovered target evidence identifier set is empty, the missing evidence compensation generation mechanism is not triggered.

[0153] The reorganization of the declaration units in the deleted declaration unit set according to the order of the nursing electronic record drafts specifically includes:

[0154] In a specific embodiment, the mechanism for generating compensation for missing evidence includes:

[0155] The evidence fragment content corresponding to the target evidence identifier is retrieved through a bidirectional index table. The supplementary declaration unit is generated using the target evidence identifier and the corresponding evidence fragment content as input information sources.

[0156] Obtain the set of evidence identifiers for the supplementary declaration unit as the supplementary evidence identifier set, and check whether the supplementary evidence identifier set includes the target evidence identifier input when generating the supplementary declaration unit;

[0157] If the check result indicates that the supplementary declaration unit is not included, delete the unit.

[0158] When the inspection result indicates inclusion, a supplementary statement unit will be added to the draft of the electronic nursing record;

[0159] The bidirectional consistency check was re-performed on the draft of the nursing electronic record after the addition of supplementary declaration units to obtain the updated set of uncovered target evidence identifiers.

[0160] like If so, the mechanism for generating compensation for missing evidence will end;

[0161] Otherwise, if Then the mechanism for generating compensation for missing evidence will end, and will Add to the list of tasks requiring manual intervention;

[0162] otherwise, ,Will Defined as Repeated execution of the missing evidence compensation generation mechanism;

[0163] in, The set of uncovered target evidence identifiers obtained after the k-th round of bidirectional consistency verification is obtained by re-performing the bidirectional consistency verification after the k-th round of compensation generation and the addition of supplementary declaration units. , The cardinality of the uncovered target evidence identifier set, It is an empty set;

[0164] The declaration units in the nursing electronic record draft at the end of the missing evidence compensation generation mechanism are reorganized in the order of the nursing electronic record draft to obtain the compensation processing draft.

[0165] Specifically, before initiating the missing evidence compensation generation mechanism, a set of uncovered target evidence identifiers is obtained based on bidirectional consistency verification, and the initial set of uncovered target evidence identifiers is defined as the set of uncovered target evidence identifiers for the first iteration. Using the set of uncovered target evidence identifiers for the first iteration as the input object for compensation generation, for each target evidence identifier in the set of uncovered target evidence identifiers for the first iteration, the corresponding evidence fragment content is retrieved through a bidirectional index table. A supplementary declaration unit is generated using the target evidence identifier and the corresponding evidence fragment content as the input information source. After the system appends the supplementary declaration unit to the nursing electronic record draft, it re-executes the bidirectional consistency verification to obtain the updated set of uncovered target evidence identifiers after the first iteration. The updated set of uncovered target evidence identifiers is used as the set of uncovered target evidence identifiers for the next iteration and enters the subsequent judgment process.

[0166] It should be noted that the generation of electronic nursing records has been transformed from the traditional generation of a seemingly reasonable text into a process of computable consistency verification and convergent compensation generation based on a closed set of evidence. The two types of risks that are most difficult to automate in nursing record quality control are also incorporated into the process control.

[0167] One type is statements without supporting evidence entering the medical record, and the other is omissions caused by the existence of key risk evidence but its absence in the record. Traditional methods of generating medical records with citations or RAG-type methods usually only impose constraints in the positive direction, that is, require the generated content to cite the retrieved documents or fragments, thereby reducing the creation of information out of thin air. However, the citation granularity is usually coarse, and more importantly, there is a lack of reverse constraints. This means that even if every sentence is cited, there may still be cases where high-risk evidence fragments such as monitoring alarms, significant abnormal vital signs, or critical treatments are not mentioned at all in the final record. In nursing settings, the tolerance for such omissions is very low.

[0168] By breaking down the above problems into two deterministic links through bidirectional consistency verification and executing them with a unified standard, consistency checks are performed on the declaration unit side. Declaration units that encounter an empty evidence identifier set or include evidence identifiers outside the closed set of evidence are directly deleted, making the content entering the subsequent verification and update links controllable at the evidence boundary. Then, the target evidence identifier is extracted from the closed set of evidence by using the forced carry-over tag as a tool, and the uncovered target evidence identifier set is dynamically generated by searching the covered evidence identifier set. In this way, the quality control point of whether key evidence is recorded, which traditionally relies on manual sampling, is transformed into an object that the system can calculate, locate, and output.

[0169] By introducing a mechanism to compensate for missing evidence, the process goes beyond simply alerting nurses to a missed item or allowing the automated system to fill in a missing paragraph. Instead, it uses the uncovered target evidence identifier as an anchor point to retrieve the corresponding evidence fragment, generates a supplementary statement unit, and enforces verification that the evidence identifier set of the supplementary statement unit must include the target evidence identifier. This avoids issues such as writing off-topic evidence, still not covering the target evidence after supplementation, or introducing new, unsubstantiated statements. Furthermore, the compensation generation is not a one-time action but forms a closed-loop iteration with bidirectional consistency verification. After each round of addition, consistency verification is re-executed and the uncovered target evidence identifier set is updated, ensuring that the compensation process converges around the determinable condition of whether the uncovered set decreases or becomes empty. If convergence fails, a list of pending manual interventions is output as an abnormal exit path, preventing infinite loops. It retains the clinical decision entry point; compared with traditional technologies, this invention reduces the probability of unfounded statements entering the final draft by deleting passive declaration units and closed set boundary constraints in terms of generation accuracy; in terms of record completeness, it moves the mandatory carry-over of key evidence to the generation stage through mandatory carry-over tags and coverage retrieval, and can trigger automatic completion; in terms of auditability, it explicitly outputs the records corresponding to which key evidence is missing in the form of a set of uncovered target evidence identifiers and links them with subsequent compensation generation and manual lists; in terms of engineering feasibility, it makes compensation generation controllable and convergent by supplementing the verification of the target evidence identifiers that the declaration unit must cover and the iteration termination conditions. Thus, it achieves a more rigorous, closed-loop, and more compliant process optimization with the logic of nursing document accountability and quality control than the traditional solution that only makes citation constraints or only makes missing item prompts.

[0170] In a specific embodiment, step S3 specifically includes:

[0171] Step S31: According to the preset type rules, generate a declaration type mark for the declaration unit to be checked in the compensation processing draft. The declaration type includes determining the declaration type and judging the declaration type.

[0172] Step S32: Generate routing results based on the declaration type marker for the declaration unit to be verified. The routing results include rule verification routes and adversarial verification routes, specifically including:

[0173] When the declaration type is marked as a definitive declaration type, the generated route result is a rule-verified route;

[0174] When the declaration type is marked as a judgment declaration type, the generated route result is a route that resists verification;

[0175] Specifically, the preset type rules are used to generate a declaration type label based on the text content of the declaration unit to be examined and the evidence fragment content corresponding to the set of evidence identifiers it carries; specifically, these include rules for identifying fields to be verified, rules for determining field types, and rules for the availability of evidence fields.

[0176] The field identification rule is used to identify whether there is a field to be verified in the text content of the claim unit to be tested. The fields to be verified include numerical values, time, dosage, frequency, and status fields that can be represented in a structured way. The field type determination rule is used to determine the field type to which the identified field to be verified belongs, such as vital signs field type, drug administration field type, time field type, and status field type. The evidence field availability rule is used to determine whether there is an evidence field of the same type that can be used for verification in the evidence fragment content corresponding to the claim unit to be tested.

[0177] Based on preset type rules, a declaration type tag is generated for each declaration unit to be checked in the compensation processing draft, specifically including:

[0178] Obtain the text content of the declaration unit to be verified and identify whether there are fields to be checked;

[0179] When no field to be checked is identified, the system sets the declaration type flag of the declaration unit to be checked to the judgment declaration type;

[0180] When a field to be verified is identified, the system determines the field type of the field to be verified and retrieves the evidence fragment content corresponding to the evidence identifier set carried by the declaration unit to be verified through a two-way index table.

[0181] Determine if there is an evidence field of the same type in the evidence fragment that can be used for verification:

[0182] When there are evidence fields available for verification, the system will set the declaration type flag to determine the declaration type;

[0183] When no evidence field is available for verification, the system sets the declaration type flag to determine the declaration type.

[0184] Step S33: When the routing result is a rule-verified route, the rule verification is performed on the declaration unit to be verified to obtain a ternary judgment result. The ternary judgment result includes only supporting declaration units, only supporting counterfactual declaration units, and insufficient evidence.

[0185] The rule verification processing logic includes:

[0186] Extract verifiable fields from the declaration unit to be inspected, extract evidence fields of the same type as the verifiable fields from the evidence fragment content set, and perform field consistency verification according to the preset verification rules;

[0187] When the verifiable field matches the evidence field, the output ternary judgment result is "only supports declaration unit".

[0188] When the evidence field exists but is inconsistent with the verifiable field, the output ternary judgment result is that only the counterfactual declaration unit is supported.

[0189] When it is impossible to extract evidence fields of the same type as the verifiable fields, the output ternary judgment result is insufficient evidence;

[0190] Specifically, in the evidence fragment content set, evidence fields of the same type as the verifiable fields are extracted. Field consistency verification involves extracting at least one verifiable field from the text content of the statement unit to be tested. A verifiable field is a field that can be directly located and its value compared within the evidence fragment content set. A verifiable field includes a field type and a field value. The field type indicates the clinical meaning category to which the field belongs. Field types include at least one of the following: vital signs field type, medication administration field type, laboratory test field type, intake / output field type, time field type, and enumeration field type.

[0191] Obtain the set of evidence fragments corresponding to the claim unit to be examined, and perform field alignment operations on the set of evidence fragments. The field alignment operation includes locating evidence fields of the same type in the set of evidence fragments based on the field type of the verifiable fields, thus obtaining evidence fields of the same type.

[0192] The field alignment operation addresses the question of which type of evidence field in the evidence fragment content set should be compared to a verifiable field in the unit of evidence to be examined. If an evidence field of the same type as the verifiable field cannot be located in the evidence fragment content set, the rule verification result for the unit of evidence to be examined is output as insufficient evidence.

[0193] Once field alignment is complete and similar evidence fields are obtained, a field consistency check is performed between the values ​​of the verifiable fields and the values ​​of the similar evidence fields. The field consistency check includes comparing the values ​​of the two fields according to preset equivalence rules to determine whether they are consistent or conflicting.

[0194] The preset equivalence rules are used to determine whether the verifiable field values ​​in the declaration unit to be tested are consistent with the values ​​of the same type of evidence fields in the evidence fragment content. These rules include unit and format normalization rules, numerical field equivalence rules, time field equivalence rules, status field equivalence rules, drug and treatment name equivalence rules, and unverifiable rules.

[0195] The unit and format normalization rule is to normalize the values ​​of both fields before comparison, convert the units to a unified unit, and then convert the numerical strings to a standard numerical format.

[0196] Numerical field equivalence rules apply to heart rate, blood pressure, body temperature, blood oxygen saturation, intake and output, and drug dosage. Specifically, they include determining consistency if the values ​​on both sides are equal or within the allowed preset tolerance.

[0197] Inconsistent judgments exceeding tolerance;

[0198] The preset tolerance range is configured and stored by the system for different field types.

[0199] The time field equivalence rule applies to occurrence time, administration time, and retest time, specifically including cases where the two times are completely consistent and cases where the preset time window allows for consistency.

[0200] Inconsistent judgments due to exceeding the time window;

[0201] The preset time window is pre-configured and stored by the system.

[0202] The equivalence rule for status fields applies to presence / absence, whether / not present, not yet present, awake / sleepy, and stable / unstable. Specifically, it includes performing terminology normalization on the values ​​of both fields, mapping synonymous expressions to unified standard terms, and ensuring consistency if the standard terms are the same, and inconsistency if the standard terms are different.

[0203] The synonym mapping table is pre-configured and stored by the system.

[0204] The equivalence rule between drug and treatment names applies to drug administration and nursing care, specifically including that if the standard names or standard codes are the same, they are consistent; otherwise, they are inconsistent.

[0205] The rule for "cannot be verified" is that when it is impossible to locate an evidence field of the same type as the verifiable field in the evidence fragment content, or when it is impossible to complete the normalization process of the field value, the system outputs "cannot be verified" and outputs the ternary judgment result of the declaration unit as "insufficient evidence" under the rule verification route.

[0206] Pre-configured tolerances, preset time windows, and synonym mapping tables are pre-configured and stored.

[0207] The preset tolerance is preferably a set of allowable difference rules corresponding to the field type, used to determine the allowable deviation range of numerical fields during comparison. The preset tolerance is preferably stored in a field type-allowable deviation correspondence relationship, with field types including heart rate, blood pressure, body temperature, blood oxygen saturation, intake / output, and drug dosage. The preset time window is preferably a set of allowable time deviation rules corresponding to the time field type, used to determine the allowable deviation range of time fields during comparison. The preset time window is preferably stored in a time field type-allowable time deviation correspondence relationship, with time field types including occurrence time, administration time, retest time, and recording time. The synonym mapping table is preferably a set of mapping rules for terminology normalization, used to map enumeration items or state items with different expressions but semantically equivalent meanings to unified standard items. The synonym mapping table is preferably stored in a synonym expression-standard item correspondence relationship, with state items including presence / absence, yes / no, not yet, awake / drowsy, and stable / unstable.

[0208] When the routing result is a route that resists verification, the unit to be verified is subjected to resist verification to obtain a ternary determination result;

[0209] By combining the text content of the statement unit to be tested with the statement type tag, a counterfactual statement unit is generated, and a triple is constructed. The triple includes a set of evidence fragments, the statement unit to be tested, and the counterfactual statement unit.

[0210] Given a set of evidence fragments, the evidence support judgment is performed on the unit of statement to be examined and the counterfactual unit, respectively. The processing expression is as follows:

[0211] ;

[0212] ;

[0213] in, To determine whether the set of evidence fragments supports the declaration unit. As a result, For the unit of declaration to be tested, As a counterfactual statement unit, , For supportive decision values, , 1 indicates support, 0 indicates no support;

[0214] The ternary decision result is output based on the support decision value, specifically including:

[0215] when =1 and When the value is 0, the output ternary determination result is "only supports declaration units";

[0216] when =0 and When the value is 1, the output ternary determination result is that only counterfactual declaration units are supported;

[0217] Otherwise, the ternary determination result is insufficient evidence;

[0218] Specifically, generating counterfactual statement units by combining the text content of the statement unit to be tested with the statement type tag includes:

[0219] Obtain the text content of the declaration unit to be tested and the declaration type tag corresponding to the declaration unit to be tested; extract assertion elements from the text content of the declaration unit to be tested. The assertion elements include at least subject elements and statement elements. The statement elements include event or state elements and attribute elements corresponding to the event or state elements. Among them, the attribute elements include numerical value, time, frequency, degree, location and state.

[0220] Based on the declaration type tag, a preset counterfactual generation rule corresponding to the declaration type tag is selected. The preset counterfactual generation rule is used to indicate the way to perform conflict transformation on the assertion element, specifically including existence / non-existence transformation, completed / incomplete transformation, affirmative / negative expression transformation, normal / abnormal transformation, degree enhancement / degree reduction transformation, and state normal conflict transformation.

[0221] The system performs counterfactual transformations on assertion elements according to preset counterfactual generation rules. Counterfactual transformations include converting existential statements to non-existent statements, completed statements to incomplete statements, affirmative statements to negative statements, normal statements to anomalous statements, statements that enhance intensity to statements that weaken intensity, and states to states that conflict with them. The counterfactual transformations are then reorganized into counterfactual statement unit text content, generating counterfactual statement units that are identical to the statement unit to be tested in terms of subject elements and statement type. An opposition check is performed on the counterfactual statement unit and the statement unit to be tested to confirm a conflict between them. If the opposition check is not satisfied, the system re-executes the selection and counterfactual transformation of the preset counterfactual generation rules until a counterfactual statement unit that satisfies the opposition check is generated.

[0222] Here, whether the evidence supports a certain statement is abstracted into a decision function Entail, where E(u) is the set of evidence fragments bound to the statement unit to be tested, and u is the original statement to be verified. For declarations that conflict with the semantics of u. This indicates that, considering only evidence E(u), we determine whether the evidence supports the truth of u. If a=1, the evidence supports u; if a=0, the evidence does not support u. This refers to determining whether the evidence supports the claim when only considering evidence E(u). If true, then b=1 is the supporting evidence. b=0 indicates that the evidence does not support the claim. a and b are the binary outputs of the two evidentiary support judgments, respectively for the original statement and the counterfactual statement.

[0223] Insufficient evidence applies to all other cases, including two common scenarios:

[0224] Since a=0 and b=0, the evidence is insufficient for both sides, indicating that the original text was unclear.

[0225] If a=1 and b=1, the evidence supports both sides, which usually means that there is a conflict of evidence or the statement is too broad.

[0226] Step S34: Record the declaration unit identifier, the declaration type tag of the declaration unit, the routing result, the evidence identifier set, the evidence fragment content set, and the ternary judgment result as judgment log entries, and bind the judgment log entries to the declaration unit identifier.

[0227] It should be noted that common practices in traditional technologies include generating nursing records directly without verification, allowing the model to perform self-checks on its output, or using similarity or keyword matching for weak verification. These methods generally have two drawbacks. First, when faced with a large number of judgmental statements with sensitive wording in nursing records, simple similarity verification cannot determine whether there is an overstatement. Furthermore, model self-checks are inherently prone to confirmation bias, easily misjudging statements that seem reasonable but for which the evidence is not explicitly stated as correct. Second, when faced with objective numerical values ​​or fixed medical orders, if large-scale model semantic verification or adversarial verification is still used, it will cause unnecessary verification overhead and result in unstable interpretation.

[0228] After the compensation processing draft is formed, this invention generates a declaration type tag for each declaration unit to be tested according to preset type rules, and generates a routing result accordingly. Declaration units with determined declaration types enter the rule verification route, and a ternary judgment result is output based on the consistency of verifiable field extraction, same-type evidence field location, and preset verification rules. This transforms directly verifiable objective fact verification into a deterministic field-level comparison, thereby reducing reliance on semantic models and improving the consistency of verification criteria. Declaration units with determined declaration types enter the adversarial verification route. A counterfactual declaration unit is generated using the text content of the declaration unit and the declaration type tag, and a triple containing a set of evidence fragments, the declaration unit to be tested, and the counterfactual declaration unit is constructed. Under the given set of evidence fragments, evidence support judgments are performed on the declaration unit to be tested and the counterfactual declaration unit respectively, obtaining support judgment values ​​a and b. Then, (a, b) is mapped to the ternary judgment result. The key to this design is that it does not ask whether the statement resembles evidence, but rather requires simultaneously judging whether the evidence supports the original statement and whether the evidence supports the opposing statement within the same evidence boundary, and using exclusive groups... The convergence relation converges the verification results into an executable ternary output, thus being better at identifying the most common and difficult-to-automate risks in nursing records. It addresses situations where evidence is cited but the wording is exaggerated or inferences exceed the boundaries of the evidence. For example, if the evidence fragment only indicates one out-of-bed activity today, and the statement to be tested is written as "free movement," then when generating a counterfactual statement of "limited movement" under the adversarial verification route and making an evidence support judgment, it often falls into the category of insufficient evidence. This prompts subsequent update operations to delete the statement or rewrite it to a more evidence-based description. Therefore, compared to traditional technologies, this invention achieves a higher level of verification strategy. The system employs an adaptive distribution mechanism that uses rule-based verification for objective fields and adversarial verification for judgment statements. This avoids the cost waste and instability caused by using a single verification method to cover all claims. In terms of verification logic, the system uses a ternary mapping of counterfactual comparison and supporting judgment values ​​to elevate the semantic verification of nursing records from subjective scoring or self-confirmation to a decidable and convergent process output. In terms of result utilization, the system provides a referenceable structured basis for subsequent updates, final draft review, and manual intervention by binding judgment log entries. This achieves stricter in-process control over the factual boundaries of nursing records and a more auditable verification loop.

[0229] In a specific embodiment, step S4 specifically includes:

[0230] Step S41: Perform update operations on the declaration units to be tested in the compensation processing draft according to the ternary determination results. The update operations include retention, deletion and downgrade rewriting.

[0231] Step S42: When the ternary determination result is that only the declaration unit is supported, the declaration unit to be tested is retained unchanged;

[0232] When the ternary determination result indicates that only counterfactual statement units are supported, the statement unit to be tested is deleted and written to the determination log entry;

[0233] When the ternary triad result indicates insufficient evidence, evidence identification is corrected to obtain an updated ternary triad result. An update operation is then performed based on the updated ternary triad result, specifically including:

[0234] When the updated ternary determination result is that only the declaration unit is supported, the declaration unit is retained and its evidence identifier set is updated to the corrected evidence identifier set.

[0235] When the updated ternary determination result is not "only supports declaration units", the declaration unit is downgraded and rewritten to generate an alternative declaration unit.

[0236] When the replacement declaration unit is successfully generated, the declaration unit is replaced with the replacement declaration unit.

[0237] If the generation of a replacement declaration unit fails, the declaration unit is deleted.

[0238] The generation of alternative declaration units specifically includes:

[0239] The set of evidence fragments to be downgraded and rewritten is retrieved from the set of evidence identifiers of the declaration unit to be downgraded and rewritten by the bidirectional index table. The set of evidence fragments to be downgraded and rewritten is sorted according to the time order of the evidence fragments. The sorted evidence fragments to be downgraded and rewritten are then concatenated to generate the alternative declaration unit.

[0240] Step S43: Reorganize all the declaration units processed in S42 according to the order of the nursing electronic record draft to obtain the final version of the nursing electronic record;

[0241] Step S44: Perform a final balance check on the final draft of the electronic nursing record;

[0242] Extract the target evidence identifier with the mandatory carry-over label 1 from the closed set of evidence; retrieve the evidence identifier set of the statement unit to be examined based on the final draft of the nursing electronic record.

[0243] If at least one set of evidence identifiers includes the target evidence identifier, then the target evidence identifier is marked as covered.

[0244] If no evidence identifier set contains the target evidence identifier, the retrieval is deemed a failure, and the target evidence identifier is added to the manual intervention pending list.

[0245] Step S45: Output the list of pending manual interventions, the final draft of the electronic nursing record, and the judgment log entries;

[0246] Step S46: Receive manual intervention instructions, process the target declaration unit based on the manual intervention instructions, and output the final draft of the nursing electronic record after manual intervention;

[0247] The manual intervention instructions include restoration, replacement, and maintenance operations for the target declaration unit.

[0248] Specifically, during the final draft balancing review, if the system fails to find any final draft declaration unit's evidence identifier set that includes a specific target evidence identifier, the system determines that the target evidence identifier retrieval has failed and adds the target evidence identifier to the manual intervention pending list. This list generation and writing operation is completed automatically by the system. Step S45, which outputs the manual intervention pending list, the final draft of the nursing electronic record, and the judgment log entries, is a system output operation. The system displays these contents on the interactive interface for subsequent generation and submission of manual intervention instructions; at this stage, the system does not make any further manual modifications to the final draft content.

[0249] After reviewing the pending manual intervention list, the final draft of the electronic nursing record, and the judgment log entries in the interactive interface, nursing staff select and submit manual intervention instructions for the target declaration unit. Manual intervention instructions include recovery operations, replacement operations, and maintenance operations. This selection and submission behavior constitutes manual intervention.

[0250] Step S46 involves receiving manual intervention instructions, processing the target declaration unit based on the manual intervention instructions, and outputting the final draft of the nursing electronic record after manual intervention. The system performs the following operations: writes, replaces, or leaves the final draft unchanged according to the manual intervention instructions, and records the manual intervention log entries.

[0251] In a specific embodiment, the evidence identification correction includes:

[0252] Extract the text semantic features of the statement unit to be tested, and traverse the text semantic features of each evidence fragment in the closed set of evidence. Calculate the semantic relevance score, arrange the evidence identifiers in descending order of semantic relevance score, and extract the first N evidence identifiers to form a candidate evidence identifier set, where N is a positive integer.

[0253] Select at least one candidate evidence identifier from the candidate evidence identifier set to form a modified evidence identifier set, construct the modified evidence identifier set, update the evidence fragment content set according to the modified evidence identifier set, and obtain the modified evidence fragment content set;

[0254] The corrected set of evidence identifiers is used to replace the set of evidence identifiers, and the corrected set of evidence fragment content is used to replace the set of evidence fragment content. Corresponding route verification is then performed to obtain the updated ternary judgment result.

[0255] Specifically, extracting the textual semantic features of the declaration unit to be tested includes obtaining the text content of the declaration unit; performing text normalization processing on the text content, which includes removing irrelevant symbols, unifying capitalization, normalizing synonyms, and normalizing unit expressions; inputting the normalized text into a preset text encoder, and outputting a fixed-dimensional vector as the textual semantic features of the declaration unit to be tested. The preset text encoder is preferably a general sentence vector model;

[0256] The calculation of semantic relevance score specifically includes obtaining the text semantic feature vector of the statement unit to be tested and the text semantic feature vector of the content of each evidence fragment in the closed set of evidence, calculating the cosine similarity between the two vectors, and using the cosine similarity as the semantic relevance score between the statement unit to be tested and the content of the evidence fragment.

[0257] The optimal value for N is 10. A value that is too small may miss correct evidence markers, while a value that is too large will increase the processing workload for verification and updates.

[0258] In a specific embodiment, step S45 specifically includes:

[0259] If the manual intervention instruction is a recovery operation, the deleted and replaced declaration units corresponding to the target declaration unit will be written into the final version of the nursing electronic record and used as the declaration unit after the manual intervention.

[0260] If the manual intervention instruction is a replacement operation, then the corresponding declaration unit in the final draft of the nursing electronic record will be replaced with the replacement text provided by the manual intervention instruction;

[0261] If the manual intervention instruction is to maintain the operation, then the final version of the electronic nursing record remains unchanged.

[0262] Specifically, a deleted declaration unit refers to a declaration unit that, after being deleted during the update operation, will no longer be written into the final version of the electronic nursing record.

[0263] When a nursing staff issues a recovery operation and the recovery target is a deleted declaration unit, the deleted declaration unit corresponding to the target declaration unit identifier in the deletion history is located according to the target declaration unit identifier in the manual intervention instruction; the deleted declaration unit is written into the final version of the nursing electronic record and used as the declaration unit after manual intervention; the manual intervention log entry corresponding to this recovery operation is recorded, and the manual intervention log entry is bound and stored with the target declaration unit identifier.

[0264] After the system performs a downgrade rewrite, the original declaration unit is replaced with an alternative declaration unit. The original declaration unit is no longer displayed in the final text but is saved in the replacement history.

[0265] When a nursing staff member issues a recovery operation and the recovery target is a replaced declaration unit, the replaced declaration unit corresponding to the target declaration unit identifier in the replacement history is located according to the target declaration unit identifier in the manual intervention instruction; the replaced declaration unit replaces the current corresponding position of the substitute declaration unit in the final draft of the nursing electronic record, and the replaced content is used as the declaration unit after manual intervention; the manual intervention log entry corresponding to this recovery operation is recorded, and the manual intervention log entry is bound and stored with the target declaration unit identifier.

[0266] Example 2, refer to Figure 2 This is another embodiment of the present invention, which differs from the first embodiment in that it provides an automatic generation system for nursing electronic records, including a data acquisition module, a verification module, a processing module and an output module;

[0267] The acquisition module is used to segment the unified raw data and assign unique evidence identifiers to obtain evidence fragments, and encapsulate all evidence fragments into a closed set of evidence.

[0268] The verification module is used to generate a draft of the nursing electronic record carrying evidence identifiers based on the closed set of evidence, and to break it down into declaration units;

[0269] The closed set of evidence and the declaration unit are subjected to bidirectional consistency verification. An exception of mandatory evidence not being covered is handled and a mechanism for generating missing evidence compensation is implemented. The compensation processing draft is then reorganized.

[0270] The processing module is used to preset routing rules, distribute the declaration units to be verified in the compensation processing draft to different verification routes for verification, obtain the ternary judgment result and record the judgment log entries;

[0271] The output module is used to update the declaration unit according to the ternary judgment result, obtain the final draft of the nursing electronic record, perform final draft balancing and verification, and output the final draft of the nursing electronic record after manual intervention after updating with manual intervention instructions.

[0272] This invention establishes the generation of electronic nursing records based on evidence identification constraints and bidirectional consistency verification. It automatically triggers compensatory generation upon discovering uncovered critical evidence, continuing until coverage conditions are met or manual intervention is initiated. The generated content is linked one-to-one with evidence fragments, facilitating traceability and auditing. Declarations with missing evidence identification or illegal citations are deleted, reducing the risk of unprovoked fabrication. The mandatory carry-over tag system retrieves the coverage of critical evidence and outputs the uncovered set, reducing omissions of high-risk events. Compensatory generation and duplicate verification form a closed loop, improving record integrity and providing a terminateable exit mechanism. Based on type routing and counterfactual verification, a ternary judgment result is output, reducing inferences and overly strong wording. Combined with final draft review and manual instruction updates, the final draft is controllable, and records of manual decisions are preserved.

[0273] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0274] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A method for automatically generating electronic nursing records, characterized in that, Includes the following steps: Step S1: The raw data after unification is segmented and assigned a unique evidence identifier to obtain evidence fragments, and all evidence fragments are encapsulated into a closed set of evidence. Step S2: Generate a draft of the nursing electronic record carrying evidence identifier based on the closed set of evidence, and break it down into declaration units; The closed set of evidence and the declaration unit are subjected to bidirectional consistency verification. An exception of mandatory evidence not being covered is handled and a mechanism for generating missing evidence compensation is implemented. The compensation processing draft is then reorganized. Step S3: Based on the preset type rules, the declaration units to be verified in the compensation processing draft are distributed to different verification routes for verification, and the ternary judgment results are obtained and the judgment log entries are recorded. Step S4: Update the declaration unit according to the ternary judgment result to obtain the final draft of the nursing electronic record and perform final draft balancing and verification. After updating with the manual intervention instruction, output the final draft of the nursing electronic record after manual intervention.

2. The method for automatically generating electronic nursing records as described in claim 1, characterized in that, Step S1 specifically includes: Step S11: Obtain the raw data and perform unified processing on the raw data. The unified processing includes time alignment, unit alignment, and preservation of text transcription results. The raw data includes monitoring data, medical orders, voice interaction data, and manually entered data; The monitoring data includes heart rate, blood pressure, and blood oxygen saturation; Step S12: Divide the unified raw data into evidence fragments, assign a unique evidence identifier to each evidence fragment, and establish a bidirectional index table between the evidence identifier and the content of the evidence fragment through a key-value pair mapping table. Step S13: Evaluate each piece of evidence using preset risk rules; Set the mandatory carry-over label for evidence fragments that meet the preset risk rules to 1; Set the mandatory carry-over label of evidence fragments that do not comply with the preset risk rules to 0; Step S14: Encapsulate all evidence fragments into a closed set of evidence, wherein the evidence fragments include evidence identifiers and mandatory carry-over label fields.

3. The method for automatically generating electronic nursing records as described in claim 2, characterized in that, Step S2 specifically includes: Step S21: Retrieve the evidence fragment content corresponding to the evidence identifier in the closed set of evidence through the bidirectional index table, and perform anchor point constraint operation on the closed set of evidence based on the evidence fragment content to obtain the nursing electronic record draft, wherein the nursing electronic record draft carries the evidence identifier. Step S22: Divide the draft of the nursing electronic record into declaration units according to the sentence segmentation, take the text content in the draft of the nursing electronic record as the text content of the declaration unit, and extract the evidence identifier set of each declaration unit in the draft of the nursing electronic record. Step S23: Perform a two-way consistency check between the closed set of evidence and the declaration unit to obtain a draft of the compensation process; Step S24: Using a bidirectional index table, retrieve the evidence fragment content corresponding to the evidence identifier set carried by the statement unit to be tested in the compensation processing draft, and obtain the evidence fragment content set corresponding to the statement unit to be tested.

4. The method for automatically generating electronic nursing records as described in claim 3, characterized in that, Bidirectional consistency verification specifically includes: Obtain the set of evidence identifiers for the claim units of the draft nursing electronic records, and perform a claim unit consistency check: When it is detected that the evidence identifier set of the declaration unit is empty, or that any evidence identifier in the evidence identifier set does not belong to the closed set of evidence, the declaration unit is deleted; Otherwise, retain the declaration unit; Get the collection of declaration units after deletion; By summing the evidence identifier sets of the declared units in the deleted set of declared units, we obtain the covered evidence identifier set; Extract the evidence identifier with the mandatory carry-over label 1 from the closed set of evidence as the target evidence identifier, and then perform a target evidence identifier retrieval in the covering evidence identifier set: When the target evidence identifier is found to be included in the set of covered evidence identifiers, the target evidence identifier is marked as successfully retrieved. When the target evidence identifier is not found in the covered evidence identifier set, the target evidence identifier is determined to be an abnormal forced evidence not covered, recorded in the uncovered target evidence identifier set, and the missing evidence compensation generation mechanism is executed. When the search is completed and the target evidence identifier set is empty, the claim units in the deleted claim unit set are reorganized in the order of the nursing electronic record draft to obtain the compensation processing draft.

5. The method for automatically generating electronic nursing records as described in claim 4, characterized in that, The mechanism for generating compensation for missing evidence includes: The evidence fragment content corresponding to the target evidence identifier is retrieved through a bidirectional index table. The supplementary declaration unit is generated using the target evidence identifier and the corresponding evidence fragment content as input information sources. Obtain the set of evidence identifiers for the supplementary declaration unit as the supplementary evidence identifier set, and check whether the supplementary evidence identifier set includes the target evidence identifier input when generating the supplementary declaration unit; If the check result indicates that the supplementary declaration unit is not included, delete the unit. When the inspection result indicates inclusion, a supplementary statement unit will be added to the draft of the electronic nursing record; The bidirectional consistency check was re-performed on the draft of the nursing electronic record after the addition of supplementary declaration units to obtain the updated set of uncovered target evidence identifiers. like If so, the mechanism for generating compensation for missing evidence will end; Otherwise, if Then the mechanism for generating compensation for missing evidence will end, and will Add to the list of tasks requiring manual intervention; otherwise, ,Will Defined as Repeated execution of the missing evidence compensation generation mechanism; in, The set of uncovered target evidence identifiers obtained after the k-th round of bidirectional consistency verification is obtained by re-performing the bidirectional consistency verification after the k-th round of compensation generation and the addition of supplementary declaration units. , The cardinality of the uncovered target evidence identifier set, It is an empty set; The declaration units in the nursing electronic record draft at the end of the missing evidence compensation generation mechanism are reorganized in the order of the nursing electronic record draft to obtain the compensation processing draft.

6. The method for automatically generating electronic nursing records as described in claim 5, characterized in that, Step S3 specifically includes: Step S31: According to the preset type rules, generate a declaration type tag for the declaration unit to be checked in the compensation processing draft. The declaration type includes determining the declaration type and judging the declaration type. Step S32: Generate routing results based on the declaration type marker for the declaration unit to be verified. The routing results include rule verification routes and adversarial verification routes, specifically including: When the declaration type is marked as a definitive declaration type, the generated route result is a rule-verified route; When the declaration type is marked as a judgment declaration type, the generated route result is a route that resists verification; Step S33: When the routing result is a rule-verified route, the rule verification is performed on the declaration unit to be verified to obtain a ternary judgment result. The ternary judgment result includes only supporting declaration units, only supporting counterfactual declaration units, and insufficient evidence. The rule verification processing logic includes: Extract verifiable fields from the declaration unit to be inspected, extract evidence fields of the same type as the verifiable fields from the evidence fragment content set, and perform field consistency verification according to the preset verification rules; When the verifiable field matches the evidence field, the output ternary judgment result is "only supports declaration unit". When the evidence field exists but is inconsistent with the verifiable field, the output ternary judgment result is that only the counterfactual declaration unit is supported. When it is impossible to extract evidence fields of the same type as the verifiable fields, the output ternary judgment result is insufficient evidence; When the routing result is a route that resists verification, the unit to be verified is subjected to resist verification to obtain a ternary determination result; By combining the text content of the statement unit to be tested with the statement type tag, a counterfactual statement unit is generated, and a triplet is constructed. The triplet includes a set of evidence fragments, the statement unit to be tested, and the counterfactual statement unit. Given a set of evidence fragments, the evidence support judgment is performed on the unit of statement to be examined and the counterfactual unit, respectively. The processing expression is as follows: ； ； in, To determine whether the set of evidence fragments supports the declaration unit. As a result, For the unit of declaration to be tested, As a counterfactual statement unit, , For supportive decision values, , 1 indicates support, 0 indicates no support; The ternary decision result is output based on the support decision value, specifically including: when =1 and When the value is 0, the output ternary determination result is "only supports declaration units"; when =0 and When the value is 1, the output ternary determination result is that only counterfactual declaration units are supported; Otherwise, the ternary determination result is insufficient evidence; Step S34: Record the declaration unit identifier, the declaration type tag of the declaration unit, the routing result, the evidence identifier set, the evidence fragment content set, and the ternary judgment result as judgment log entries, and bind the judgment log entries to the declaration unit identifier.

7. The method for automatically generating electronic nursing records as described in claim 6, characterized in that, Step S4 specifically includes: Step S41: Perform update operations on the declaration units to be tested in the compensation processing draft according to the ternary determination results. The update operations include retention, deletion and downgrade rewriting. Step S42: When the ternary determination result is that only the declaration unit is supported, the declaration unit to be tested is retained unchanged; When the ternary determination result indicates that only counterfactual statement units are supported, the statement unit to be tested is deleted and written to the determination log entry; When the ternary triad result indicates insufficient evidence, evidence identification is corrected to obtain an updated ternary triad result. An update operation is then performed based on the updated ternary triad result, specifically including: When the updated ternary determination result is that only the declaration unit is supported, the declaration unit is retained and its evidence identifier set is updated to the corrected evidence identifier set. When the updated ternary determination result is not "only supports declaration units", the declaration unit is downgraded and rewritten to generate an alternative declaration unit. When the replacement declaration unit is successfully generated, the declaration unit is replaced with the replacement declaration unit. If the generation of a replacement declaration unit fails, the declaration unit is deleted. The generation of alternative declaration units specifically includes: The set of evidence fragments to be downgraded and rewritten is retrieved from the set of evidence identifiers of the declaration unit to be downgraded and rewritten by the bidirectional index table. The set of evidence fragments to be downgraded and rewritten is sorted according to the time order of the evidence fragments. The sorted evidence fragments to be downgraded and rewritten are then concatenated to generate the alternative declaration unit. Step S43: Reorganize all the declaration units processed in S42 according to the order of the nursing electronic record draft to obtain the final version of the nursing electronic record; Step S44: Perform a final balance check on the final draft of the electronic nursing record; Extract the target evidence identifier with the mandatory carry-over label 1 from the closed set of evidence; retrieve the evidence identifier set of the statement unit to be examined based on the final draft of the nursing electronic record. If at least one set of evidence identifiers includes the target evidence identifier, then the target evidence identifier is marked as covered. If no evidence identifier set contains the target evidence identifier, the retrieval is deemed a failure, and the target evidence identifier is added to the manual intervention pending list. Step S45: Output the list of pending manual interventions, the final draft of the electronic nursing record, and the judgment log entries; Step S46: Receive manual intervention instructions, process the target declaration unit based on the manual intervention instructions, and output the final draft of the nursing electronic record after manual intervention; The manual intervention instructions include restoration, replacement, and maintenance operations for the target declaration unit.

8. The method for automatically generating electronic nursing records as described in claim 7, characterized in that, The specific amendments to evidence identification include: Extract the text semantic features of the statement unit to be tested, and traverse the text semantic features of each evidence fragment in the closed set of evidence. Calculate the semantic relevance score, arrange the evidence identifiers in descending order of semantic relevance score, and extract the first N evidence identifiers to form a candidate evidence identifier set, where N is a positive integer. Select at least one candidate evidence identifier from the candidate evidence identifier set to form a modified evidence identifier set, construct the modified evidence identifier set, update the evidence fragment content set according to the modified evidence identifier set, and obtain the modified evidence fragment content set; The corrected set of evidence identifiers is used to replace the set of evidence identifiers, and the corrected set of evidence fragment content is used to replace the set of evidence fragment content. Corresponding route verification is then performed to obtain the updated ternary judgment result.

9. The method for automatically generating electronic nursing records as described in claim 8, characterized in that, Step S45 specifically includes: If the manual intervention instruction is a recovery operation, the deleted and replaced declaration units corresponding to the target declaration unit will be written into the final version of the nursing electronic record and used as the declaration unit after the manual intervention. If the manual intervention instruction is a replacement operation, then the corresponding declaration unit in the final draft of the nursing electronic record will be replaced with the replacement text provided by the manual intervention instruction; If the manual intervention instruction is to maintain the operation, then the final version of the electronic nursing record remains unchanged.

10. An automatic nursing electronic record generation system, applied in an automatic nursing electronic record generation method as described in any one of claims 1-9, characterized in that, It includes a data acquisition module, a verification module, a processing module, and an output module; The acquisition module is used to segment the unified raw data and assign a unique evidence identifier to obtain evidence fragments, and encapsulate all evidence fragments into a closed set of evidence. The verification module is used to generate a draft of the nursing electronic record carrying evidence identifier based on the closed set of evidence, and to break it down into declaration units; The closed set of evidence and the declaration unit are subjected to bidirectional consistency verification. An exception of mandatory evidence not being covered is handled and a mechanism for generating missing evidence compensation is implemented. The compensation processing draft is then reorganized. The processing module is used to preset routing rules, distribute the declaration units to be verified in the compensation processing draft to different verification routes for verification, obtain a ternary judgment result and record the judgment log entries. The output module is used to update the declaration unit according to the ternary judgment result, obtain the final draft of the nursing electronic record, perform final draft balancing and verification, and output the final draft of the nursing electronic record after manual intervention after updating with the manual intervention instruction.

Images