TWI750513B - Insurance claim and underwriting assistance system and implementation method thereof - Google Patents

Abstract

Insurance claim and underwriting assistance system and implementation method thereof are disclosed. The present system can extract multiple keyword strings from an electronic medical record and establish the grammatical structure of the whole electronic medical record. After that, the system can compare a medical vocabulary reference table (such as medical dictionary) and a disease classification reference table (such as ICD-10) to find out the closest medical vocabularies and the closest disease standard codes, and after that the system can determine whether the disease names corresponding to the disease standard code appears in the insurance terms of the insurance policy. If there is a match, then the system can assign a first identification mark (e.g. highlight label) to the extracted keyword strings shown in the electronic medical record, and after that the system can assign the second identification mark to an auxiliary interpretation reference table corresponding to the comparison result of insurance terms, and finally integrate the electronic medical record embedded with the first identification mark and the auxiliary interpretation reference table embedded with the second identification mark into an intelligent medical record. In this way, the intelligent medical record generated by the present system can not only improve the efficiency in underwriting operation or claim settlement operation, but also reduce the possibility of omission and misjudgments in the claim settlement operation.

Description

Auxiliary system for underwriting claims and its implementation method

The present invention relates to data processing technology, in particular to a method based on natural language processing (NLP), which can perform keyword extraction and semantic analysis on medical records, and compare medical vocabulary and disease classification reference table respectively, so as to assist reviewers based on the content of medical records Quickly determine whether the insured meets the conditions for underwriting or claim settlement, and assists in determining whether the insured has a pre-existing disease or not.

Generally speaking, after a medical insurance claim case occurs, or after accepting medical insurance business, claims or underwriting personnel in the insurance industry, in order to ensure that the risk of the insured is within the acceptable range of the insurance company, will need to settle the claims. Check the various claims items in the case, or need to evaluate the insured's medical report, health notices and underwriting manuals to ensure the accuracy of the claims case or whether the insured's risk is too high, however, Since these audits are all conducted manually by underwriting or claims adjusters, and taking the audit of claims cases as an example, the claims adjusters need to spend a lot of time in the auditing process and refer to unstructured diagnostic certificates or medical records, Refer to information such as health notices, claim manuals, claims clauses, and international classification of diseases with up to 150,000 codes that record many diseases. It is even more necessary to refer to medical dictionaries, medical abbreviations dictionaries, etc., which contain many highly professional medical vocabulary data and test item names, so as to carry out correct and unbiased matching. As a result, misjudgment, missed judgment, or omission will easily occur during manual review. The insured's pre-existing medical history is not found, which may lead to the occurrence of false claims, thereby affecting the rights and interests of all insureds. In addition, it may also occur in the diagnosis certificate or medical record. Although keywords of related diseases are mentioned, the fact is The above insured does not suffer from the disease, and the underwriting or claims personnel waste a lot of time. Therefore, the conventional underwriting or claims method is not only time-consuming, but also prone to omission or misjudgment, and there is still room for improvement.

From the above, it can be seen that due to the conventional underwriting or claims personnel conducting verification work, there are still time-consuming audit work, omission of judgment, misjudgment (for example, failure to find applicable refusal diseases) or failure to find that the insured may have a history of pre-existing diseases. Therefore, how to propose an "underwriting and claims assistance system and its implementation method" that can solve the aforementioned shortcomings is a problem to be solved.

In order to achieve the above object, the present invention proposes an underwriting and claim settlement assistant system, which can generate an intelligent medical record based on an electronic medical record (medical record) for an information device to browse the intelligent medical record. The system includes a processor, a database, a second database, a keyword extraction subsystem, a keyword parsing subsystem, a data marking subsystem, an information integration subsystem and a front-end subsystem, wherein the keyword extraction subsystem can be stored in the database The electronic medical record retrieves multiple keyword strings, and the keyword strings include a medical keyword, a time string, a numerical string and a symbol character; the keyword parsing subsystem can be based on the keyword extraction subsystem For the retrieval result of the keyword string, construct the grammatical structure of the electronic medical record, analyze the time of the medical record and a semantic of the keyword string, and then compare a plurality of terms from a medical vocabulary reference table stored in the database based on the semantic meaning Medical vocabulary, and then based on the semantics and medical vocabulary, parse out at least one disease standard code from a disease classification reference table stored in the database, and then determine whether a disease name corresponding to the disease standard code matches the one stored in the database. A disease keyword string of an insurance clause in the electronic medical record to generate a clause comparison result; the data marking subsystem can assign a plurality of visible first identification marks associated with the disease name to these keyword strings in the electronic medical record, It is also used to assign a second identification mark corresponding to the item comparison result to an auxiliary interpretation reference table according to the clause comparison result; the information integration subsystem can generate an auxiliary interpretation reference table stored in the second database, and then The electronic medical record, the first identification marks, the second identification marks and the auxiliary interpretation reference table are integrated into an intelligent medical record stored in the second database, wherein the auxiliary interpretation reference table at least contains the parsed disease name, the second identification markers and disease standard codes; and when the information device is connected to the front-end subsystem through a network to successfully read the data in the second database, the front-end subsystem can provide a viewer to make the intelligent system including the auxiliary interpretation reference table The medical record is presented in the viewer executed by the information device.

In order to enable your examiners to clearly understand the purpose, technical features and effects of the present invention, the following descriptions are combined with the diagrams for illustration, please refer to.

Please refer to “FIG. 1”, which is a system architecture diagram of the present invention. The present invention proposes an underwriting and claims assistance system 10, which can generate an intelligent medical record SMR based on an electronic medical record MR for an information device 20 to browse intelligently The medical record SMR mainly includes a processor 101, a database 102, a second database 103, a keyword extraction subsystem 104, a keyword analysis subsystem 105, a data marking subsystem 106, and an information integrator The system 107 and a foreground subsystem 108 are respectively connected to the processor 101 for information, wherein: (1) The processor 101 is used to run the underwriting and claim settlement assistance system 1, access the contents of the first and second databases (102, 103), control or trigger the aforementioned multiple subsystems, and has and has logical operations, temporary The processor 101 can be, for example, a central processing unit (CPU), a virtual processing unit (vCPU), a microprocessor (MPU), a microcontroller (MCU), an application processor (AP), an embedded processor, an application-specific integrated circuit (ASIC), a tensor processing unit (TPU), a graphics processing unit (GPU), etc., but not limited thereto, Furthermore, the processor 101 itself can run on a server (not shown in the figure), and the aforementioned server can be a physical server, a server running in the form of a virtual machine (VM), or a virtual server A server running in the form of a dedicated host (Virtual Private Server); (2) Both the database 102 and the second database 103 can be a database host, and the database 102 can store a disease classification reference table ICD_T, a medical vocabulary reference table MD_T, and an insurance clause IC for the processor 101 to access (insurance clause) and electronic medical record MR; (3) The second database 103 can store an auxiliary interpretation reference table SMR_T and an intelligent medical record SMR for the processor 101 to access; and the disease classification reference table ICD_T can be an International Classification of Diseases 9th edition, 10th edition version or a combination thereof; the medical vocabulary reference table MD_T may include one or a combination of a medical dictionary and a medical abbreviation dictionary; the insurance clause IC may be an underwriting clause (or underwriting Manual), a claim clause, a claim manual and a health notice clause (also called a health notice or a health notice) one or a combination thereof; (4) The keyword extraction subsystem 104 can store a plurality of instructions, which can be triggered by the processor 101 to perform, for example, a word segmentation/tokenize process, and retrieved from the electronic medical record MR stored in the database 102 Multiple keyword strings, and these keyword strings include a medical keyword, a time string, a numerical string, and a symbol character. In addition, when the keyword extraction subsystem 104 performs word segmentation processing, it can be ignored first Punctuation that has no effect on semantics; (5) Wherein, the aforementioned medical keywords can be keywords such as a disease name, an organ name, a test name, a description phenomenon name, a description level name, and an abbreviated name. More specifically, the disease name can be, for example, For hypertension, goiter, anorexia and other diseases; the name of the organ can be, for example, the mitral valve structure, the structure of the heart, the structure of the lung, the body part and other organs; the name of the test can be, for example, echocardiography / tracing, blood gas measurement, X-ray Name of examination such as imaging procedure; descriptive name may be palpitations, married, severe, stress/compression, weight loss, skin swelling, heart murmur, sweating, etc.; descriptive name may be, for example, exacerbation, condition/morbid, complete Abbreviated names may eg be PO (oral), QN (once nightly), QD (once daily), HCVD (hypertension) Cardiovascular disease), Af (atrial fibrillation), PVCs (early ventricular contraction), OPD (outpatient), emergency (ER) and other abbreviations, but the above are only examples and are not limited to this; (6) Wherein, the aforementioned time string can be, for example, time, year, old (old), month, recent, frequent, hour, week, minute, date, same day (same day), next day (next day), but The above is only an example and is not limited to this; (7) Wherein, the aforementioned numerical string can be, for example, a test reading (such as mmHg, mm, mEqL, mg/DL, ph value, MmhG, mmol/L, %, uL, g/dL, etc., but the above are only Examples are not limited to this; (8) Wherein, the aforementioned symbol characters can be, for example, (+), (-), etc., which can indicate that the insured suffers from (+) or Not suffering from (-) the aforementioned disease name; (9) The keyword parsing subsystem 105 can store a plurality of instructions, which, after being triggered by the processor 101, can construct one or more grammars of the electronic medical record MR based on the retrieval result of the keyword string by the keyword extraction subsystem 104 structure, analyze a medical record time, and analyze a semantic of the keyword string from the aforementioned grammatical structure (for example, it can be achieved through a sentence dependency parsing), and then compare it from the medical vocabulary reference table MD_T based on the semantics. Identify multiple medical terms, and then parse out at least one disease standard code (eg, ICD Code) from the disease classification reference table ICD_T based on the semantics and the medical terms, and then determine whether a disease name corresponding to the disease standard code matches. A disease keyword string in the insurance clause IC to generate a clause comparison result; (10) Wherein, the aforementioned grammatical structure means that each descriptive sentence in the electronic medical record MR can be represented as a “Structure Tree” applied by natural language processing, and each structure tree can contain a Multiple semantic roles, such as Agent, Goal, Patient, Theme, Instrument, Location, Source, Time (Time), predicate (Predicate) and other roles; (11) Among them, the aforementioned medical record time can be a hospitalization date, a medical record preparation date, a discharge date, and a historical illness period; (12) In addition, in order to simultaneously improve the problem that the keyword extraction result or the semantic analysis result may have low accuracy due to insufficient training samples, the present invention can make the keyword extraction subsystem 104 or the keyword analysis subsystem 105 possible by letting the keyword extraction subsystem 104 or the keyword analysis subsystem 105 Read the text of a large number of electronic medical records MR, and use the statistical characteristics of the context to learn the word vector (Word Vector or Word Embedding) of each keyword string. For example, the word vector distance of "Drug" and "Abuse" is in the vector It may be close in the space, but the word vector distance of "Drug" and "smoking" may be farther in the vector space, and as more and more texts are trained, the keyword extraction subsystem 104 or the keyword parsing subsystem 105 can automatically adjust the word vector of each keyword string, solve the problem of insufficient training data for natural language processing models, and improve the abstract thinking of the system. In addition, for training word vectors, you can use training models such as CBOW and Skipgram. achieve; (13) The data labeling subsystem 106 can store a plurality of instructions, which, after being triggered by the processor 101, can assign a plurality of visible first identification marks H1 associated with the disease name to the keyword strings in the electronic medical record MD , and a second identification mark H2 corresponding to the clause comparison result can also be assigned to the auxiliary interpretation reference table SMR_T; (14) The information integration subsystem 107 can store a plurality of instructions, which, after being triggered by the processor 101, can generate the auxiliary interpretation reference table SMR_T stored in the second database 103, and then combine the electronic medical records, the first identification marks H1 , the second identification mark H2 and the auxiliary interpretation reference table SMR_T are integrated into an intelligent medical record SMR, wherein the auxiliary interpretation reference table SMR_T contains at least the disease name, the second identification mark H2 and the disease standard code that are parsed out of these; (15) The foreground subsystem 108 is available for the information device 20 to communicate with it via a network 30. When the information device 20 is connected to the foreground subsystem 108 to read the second database 103, the foreground subsystem 108 can provide the information device 20 is a viewer V, so that the intelligent medical record SMR including the auxiliary interpretation reference table SMR_T is displayed on the viewer V, so that the user (underwriting/claims adjuster) of the information device 20 can only rely on the analysis result of the intelligent medical record SMR , to review and compare the important information in the electronic medical record MR with the underwriting or claims clauses, without having to fully review the content of the electronic medical record MR; (16) In addition, the network 30 can be a public or private network, such as a wireless network (3G, 4G LTE, Wi-Fi, etc.), a wired network, a local area network (LAN), a wide area network (WA), etc.; (17) In addition, the format of the intelligent medical record SMR can be .pdf, .xls or .doc, but it is not limited to this; (18) In addition, the viewer V can be a web browser (Web Browser), a PDF viewer (PDF Viewer) or other document viewers; (19) In addition, the electronic medical record MR can use an optical character recognition (OCR) subsystem (not shown in the figure) coupled with the processor 101 to use a scanner or other optical equipment to capture the text data of a paper medical record Take digital images (ie, electronic medical record MR), and then perform automatic identification to obtain editable medical record text and layout information.

In addition, please continue to refer to FIG. 1. In a preferred embodiment of the present invention, if the medical keywords extracted by the keyword parsing subsystem 105 from a medication history of the electronic medical record MR can be obtained from the medical vocabulary reference table MD_T When the medical vocabulary belonging to "drug name" is compared, but the semantics of a specific disease is not detected from the grammatical structure of the medical keywords in the electronic medical record MR, the keyword parsing subsystem 105 can refer to the disease classification table ICD_T at this time. The disease standard code corresponding to the aforementioned drug name and the corresponding disease name are parsed to prompt the underwriting or claim settlement personnel who operate the information device 20. Although the electronic medical record MR does not explicitly mention that the patient (insured) suffers from a certain disease Specific diseases (such as high blood pressure), but it can be found from the medication history of the electronic medical record MR (such as the insured who is actually taking high blood pressure drugs on a regular basis) that the insured has a pre-existing disease of hidden high blood pressure (that is, the insured is insured while sick). possible.

Please refer to "Fig. 2", which is a flow chart of the implementation of the system of the present invention. Please refer to "Fig. 1" together with the schematic implementation diagrams of "Fig. 3" to "Fig. 7". The present invention proposes a The implementation method S of the underwriting and claim settlement assistance system may include the following steps: (1) Keyword extraction (step S10 ): a keyword extraction subsystem 104 can perform a word segmentation process to use, for example, a lexical analyzer ), the character sequence (ie, sentence, sentence) stored in the electronic medical record MR of the database 102 is divided into a word sequence composed of a plurality of words (Token), and the classification to which each word belongs is indicated, for example, one of the words It can be defined as a behavior classification (or event classification), a time expression classification, a short sentence classification, a modification classification, etc., but it is not limited to this. The electronic medical record MR captures multiple keyword strings, or executes a semantic frame parser (Parsing, such as the semantic frame parser SLING released by Google, but not limited thereto) to utilize the semantic frame (Frame Semantic Parsing) to extract the text structure of the electronic medical record MR, and present the keyword strings in the form of a Semantic frame graph, and the keyword strings include a medical keyword, a time string, and a numerical value. string and a symbol character, that is, a part of the string marked with an underline in "Figure 3", wherein the keyword extraction subsystem 104 can pre-store multiple data training sets to store completed training and some key Words already have a tag (Tag) vocabulary, and when performing word segmentation processing, the keyword extraction subsystem 104 can, for example, use "blank", ",", "-", """, ".", ":" and "," are used as separators to achieve word segmentation; (2) Grammar and semantic analysis (step S20): a keyword parsing subsystem 105 uses, for example, A syntax analyzer (also called parser) constructs one or more grammatical structures of the electronic medical record MR (for example, performing part-of-speech tagging, stemming, for example, from "leaves" → "leav"), lemmatization (e.g. from "started" → "start") and parser tree), parsing a medical record time, and using e.g. a semantic analyzer from the aforementioned grammar Structural analysis of the semantics of these keyword strings (for example, after assigning a semantic role to each word in the structure tree, it is helpful for the semantic analyzer to deduce the semantics of these keyword strings); (3) Comparison reference table ( Step S30): The keyword parsing subsystem 105 selects a medical term based on the semantics Compare multiple medical terms according to the table MD_T, and then parse out at least one disease standard code and its corresponding disease name from a disease classification reference table ICD_T based on the semantics and the medical terms, as shown in Figure 4 Based on the results of semantic analysis, medical vocabulary (such as hypertension disorder-hypertension in Figure 4), disease standard codes (such as "ICD9CM in Figure 4") are compared from the medical vocabulary reference table MD_T and the disease classification reference table ICD_T: 405.99” or “ICD10AM:I10-I15.9”) and disease name (such as “other secondary hypertension” corresponding to ICD9 in Figure 4 and “secondary hypertension” corresponding to ICD 10) (4) Compare insurance clauses (step S40): The keyword parsing subsystem 105 determines whether the disease name corresponding to the disease standard code matches a disease keyword string of an insurance clause IC to generate a clause comparison result, That is, as in "Fig. 5" based on the semantics of the electronic medical record MR, compare the disease keyword string (such as hypertension shown in Fig. 5) with the insurance clause IC such as health notification clauses (or health notification items); ( 5) Keyword marking (step S50 ): A data marking subsystem 106 assigns a plurality of first identification marks H1 that are visible to the disease name to the keyword strings in the electronic medical record MR, as shown in FIG. 6 . ” is marked with dotted lines in the form of these keyword strings, but the form of identifying the first mark H1 shown in the figure is only an example, and can also be highlighted, such as online syntax highlighting but not in this way This can be achieved with limited related technologies, and the data marking subsystem 106 can also assign at least one second identification mark H2 corresponding to the item comparison result to the auxiliary interpretation reference table SMR_T; (6) Integrate intelligent medical records (step S60): An information integration subsystem 107 generates an auxiliary interpretation reference table SMR_T, as shown in "PAGE 1" of "FIG. 7", wherein the auxiliary interpretation reference table SMR_T may include at least one disease name field and at least one insurance clause comparison The result field and the at least one disease standard code field, the contents of each field can be respectively set as the disease name, the second identification mark H2, the disease standard code and a doctor's description of the disease ( After that, the information integration subsystem 107 integrates the electronic medical record MR, the first identification marks H1, the second identification marks H2 and the auxiliary interpretation reference table SMR_T into an intelligent medical record SMR, Wherein, the auxiliary interpretation reference table SMR_T can at least include the parsed disease name, the second identification mark H2 and the disease standard code. As for other components of the intelligent medical record SMR (the first identification mark H1, the electronic medical record MR, etc.), the It is presented on other pages of the intelligent medical record SMR, so it is not shown in "Figure 7"; (7) Provide intelligent medical record inspection Viewing function (step S70 ): a foreground subsystem 108 provides a viewer V to the information device 20 , so that the intelligent medical record SMR including the auxiliary interpretation reference table SMR_T can be displayed on the viewer V.

Please continue to refer to "Fig. 8", which is another preferred embodiment (1) of the present invention, and please refer to "Fig. 1" ~ "Fig. 2" in conjunction with this embodiment and "Fig. 1" ~ The technology disclosed in FIG. 7 is similar, the main difference is that the underwriting and claim settlement assistance system 10 of this embodiment may further include a keyword correction subsystem 109 that is informationally connected to the processor 101, which is used for triggering by the processor 101. , the keyword extraction subsystem 104 can perform word segmentation processing on the keyword strings (for example, "The History !!! of @rug abuse" shown in FIG. 8 ), perform a spell check, In order to compare the most similar keyword sentences (such as "The History of Drug Abuse" shown in Figure 8) from the multiple keyword sentences stored in the database 102, and compare the meaningless or wrong keyword strings (such as "!!!" and "@" as shown in Figure 8) are removed (such as deleting "!!!") and corrected (such as correcting "@" to "d") to generate For the keyword parsing subsystem 105 to construct the grammatical structure of the electronic medical record MR, to analyze the time and semantics of the medical record, a keyword string correction result required; Step: Keyword Correction (Step S15 ): The keyword correction subsystem 109 may perform spell checking on the keyword strings retrieved after the execution of Step S10 (Keyword Extraction), so as to obtain a number of stored keywords from the database 102 . In the key sentences of the pen, the closest key sentences are compared, and the meaningless or wrong parts in the key string are removed and corrected, so as to generate step S20 (grammar and semantic analysis) for constructing the grammar of the electronic medical record MR A keyword string correction result required for structure, analysis time and semantics of medical records;

As mentioned above, please continue to refer to "Fig. 8", and please refer to "Fig. 1" ~ "Fig. 2" in conjunction. The database 102 of this embodiment can also store a lexical analyzer created by a lexical analyzer. A symbol table, for the keyword correction subsystem 109 to check whether the pattern (Pattern) to which each word (Token) belongs conforms to the symbol table for the aforementioned keyword string after word segmentation processing according to the symbol table For example, if the word is a time expression classification, check whether the expression form (lexeme) of the time string in the keyword string is expressed as a form including date, year, time or any combination thereof, and correct or shift it. remove.

Please continue to refer to "Fig. 9", which is another preferred embodiment (2) of the present invention, and please refer to "Fig. 1" ~ "Fig. 2" in conjunction with this embodiment and "Fig. 1" ~ The technology disclosed in FIG. 7 is similar, but the main difference is that, in the underwriting and claim settlement assistance system 10 of the present embodiment, the front-end subsystem 108 can also provide a searcher S coupled to the viewer V for information The device 20 defines a specified search condition (for example, specified search "hepatitis" or "liver cirrhosis") in the viewer V to search for a specified search result S_T, wherein the specified search result S_T may be a data table (Table), It may further include, for example, a specified search condition (such as hepatitis or liver cirrhosis), a clause comparison result (such as whether hepatitis or liver cirrhosis appears in the "insured person's health notification clause"), an index page number (such as the 9th P6 and P5 shown in the figure), an electronic medical record description paragraph, a medical record time (such as "1998", "2011/08" shown in Figure 9) and at least one disease standard code (such as shown in Figure 9) ICD 9, ICD 10 or a combination thereof), the aforementioned specified search condition may be the specific medical term that the keyword parsing subsystem 105 wants to compare from the medical term reference table MD_T, or the specific medical term to be parsed from the disease classification reference table The disease standard code, or the screening basis for judging whether the disease name matches the specific disease keyword string of the insurance clause, and, in a preferred embodiment of the present invention, the search engine S is also available for the information device 20 to view The device V defines a medical record time interval, and the aforementioned medical record time interval is the basis for filtering out a specific medical record time by the keyword parsing subsystem 105 based on the retrieval results of the keyword extraction subsystem 104 for these time strings; in other words, this The implementation method of the underwriting and claim settlement assistance system disclosed by the invention may further include the following steps: providing a searcher function (step S35 ): the front-end subsystem 108 provides a searcher S for the information device 20 for the information device 20 to search the viewer V defines a specified search condition and a medical record time, wherein the specified search condition is the basis for the specific medical term to be compared or the specific disease standard code to be parsed when step S30 (comparing the reference table) is executed, and also The step S40 (comparing insurance terms) may be a screening basis for determining whether it matches the specific keyword string of the disease.

Please continue to refer to "Fig. 9", and please refer to "Fig. 1" ~ "Fig. 2". The designated search results S_T in this embodiment can also be based on the order of occurrence time of each designated search disease (for example, Fig. 8). "1998"→"2011/08"→"1998"), a sequence of data points arranged in the form of, for example, a time series table (Timeline Table), to provide the underwriting or claims adjusters who review the medical records for a clearer understanding The time sequence relationship of events in the medical record helps underwriting or claims adjusters to shorten the time for reviewing unstructured texts, but is not limited to this; wherein, in this embodiment, the aforementioned time sequence table is presented, and the processor 101 can access the database Based on a time series relationship training data set stored in 102, the keyword extraction subsystem 104 uses a pre-trained (pre-trained) natural language processing model to analyze multiple character sequences (ie, given sentences) in the electronic medical record MR. It is divided into word sequences composed of multiple words (Token), and the classification of each word is respectively marked as an event (Event), a time expression (Time expressions) or an other classification, but not limited to this; , the processor 101 can then make the keyword parsing subsystem 105 use the aforementioned pre-trained natural language processing model to extract each event from the word sequence (including events, time expressions, etc.) segmented by the keyword extraction subsystem 104. A temporal relationship in a given sentence (character sequence), and the temporal relationship is classified into, for example, "before", "after" and "overlap", after which the keyword parsing subsystem 105 Then, based on the temporal relationship of each event in a given sentence (character sequence), and the time expression possible path of each event between each sentence, parse out the relationship between each event in a plurality of given sentences (character sequence). The temporal relationship can then generate one or more Temporal Graphs, so that the temporal relationship of events in each electronic medical record MR can be described by each temporal graph, which is then used as a basis for generating, for example, a temporal sequence table.

Continuing from the above, please continue to refer to "Figure 9". For the temporal relationship training data set mentioned above in this embodiment, please refer to "Evaluating temporal relations in clinical text: 2012 i2b2 Challenge; reference information: https://www .ncbi.nlm.nih.gov/pmc/articles/PMC3756273/" clinical temporal relation corpus dataset.

As mentioned above, please continue to refer to "Figure 9", and please refer to "Figure 1" ~ "Figure 2", the pre-trained natural language processing model mentioned in this embodiment can be a BERT (Bidirectional Encoder Representations from Transformers) language representative model, a Long Short Term Memory Network (LSTM) model; and, the aforementioned BERT language representative model can be a general model that has a certain understanding of natural language and is pre-trained, so the keyword The extraction subsystem 104 can use the BERT language representative model to perform two-stage transfer learning (Transfer Learning), so as to perform supervised training lexical analysis in the medical field to which the present invention is applied after the fine-tuning of parameters is completed. The task (that is, the task of converting the character sequence into the token sequence when step S10 is executed).

Continuing from the above, please continue to refer to "Fig. 9", and please refer to "Fig. 1" ~ "Fig. 2" in combination, when the keyword extraction subsystem 104 of this embodiment uses the BERT language representative model to perform the aforementioned lexical analysis For tasks, the system can use, for example, "Bert: Pre-training of deep bidirectional transformers for language understanding" published by Google AI Language; reference information can be found at: https://arxiv.org/pdf/1810.04805.pdf", and "classtransformers. BertForTokenClassification(BertPreTrainedModel); reference information can be found in: https://huggingface.co/transformers/model_doc/bert.html#bertfortokenclassification" and other technologies, but it is not limited to the models mentioned above.

Continuing from the above, please continue to refer to "Fig. 9", and please refer to "Fig. 1" ~ "Fig. 2" in combination, when the keyword parsing subsystem 105 of this embodiment uses a long short-term memory network (LSTM) model to extract For the temporal relationship of events in a given statement, for example, "LSTM-Based Model for Extracting Temporal Relations from Korean Text; reference information can be found at: https://ieeexplore.ieee.org/abstract/document/8367201" However, it is not limited to the Korean text used in this document, and the keyword parsing subsystem 105 uses a Long Short-Term Memory (LSTM) model to parse out the time sequence of each event in a plurality of given sentences For example, "Towards generating a patient's timeline: Extracting temporal relationships from clinical notes; reference information can be found at: https://www.ncbi.nlm.nih.gov/pubmed/24212118" related technologies, but not limited thereto.

Please continue to refer to "Fig. 10", which is another preferred embodiment (3) of the present invention, and please refer to "Fig. 1" ~ "Fig. 2" in conjunction with this embodiment and "Fig. 1" ~ The technology disclosed in FIG. 7 is similar, but the main difference is that, in the underwriting and claim settlement assistance system 10 of this embodiment, the front-end subsystem 108 can also provide a filter F coupled to the viewer V, and the filter F It can include a disease inspection list F_LIST (for example, including "health notification disease", "other reference information", which includes one or more screening items F_ITEM, and the screening item F_ITEM can at least correspond to each disease parsed by the keyword analysis subsystem 105 name, and each filter item F_ITEM includes an index link L1 that can lead to the relevant keyword string in the electronic medical record MR. More specifically, when the information device 20 triggers one of the filter items F_ITEM in the filter F (as shown in the figure) After the index link L1 attached to the shown “Page 4: hypertension”), the foreground subsystem 108 can guide the information device 20 to switch to the electronic medical record page marked with the keyword string “hypertension” of the first identification mark H1. Specifically, when the information device 20 triggers the index link L1 attached to one of the filter items F_ITEM in the filter F (“Page 4: Cardiomegaly” as shown in the figure), the foreground subsystem 108 can guide the information device 20 to switch Go to the electronic medical record page marked with the keyword string "Cardiomegaly" with the first identification mark H1; in other words, the implementation method S of the underwriting and claim settlement assistant system disclosed in the present invention may further include the following steps: providing a filter function (S80) : The front-end subsystem 108 provides a filter F for the information device 20. The filter F may include a disease inspection list F_LIST, and the disease inspection list F_LIST may include one or more screening items F_ITEM, and each screening item F_ITEM corresponds to step S30 ( The name of each disease parsed during the execution of the comparison reference table), and each screening item F_ITEM includes an index link L1 that can lead to the relevant keyword string in the electronic medical record MR;

As mentioned above, please continue to refer to "Fig. 10", and please refer to "Fig. 1" in combination. After the first identification mark H1 of this embodiment is triggered by the information device 20 (for example, clicked), the front-end subsystem 108 can also be activated in The keyword string "Cardiomegaly" or "Page 4: Cardiomegaly" marked with the first identification mark H1 displays an annotation information COM, wherein the annotation information COM may include an annotation information (for example, "Hypertensive Disorder Hypertension disease"), a time information, a classification information (eg "disease") and at least one disease standard code (eg, ICD9CM and ICD10CM in Figure 10), but not limited thereto.

To sum up, after the implementation of the present invention, it can at least greatly improve the review efficiency of underwriting or claim settlement personnel, reduce the time for completely reading medical records, reduce the learning cost of underwriting/claim settlement work, and reduce the possibility of misjudgment of underwriting/claim settlement. It has the beneficial effects of reducing the loss of claims due to the concealed pre-existing conditions of the insured and effectively assisting the underwriting/claims personnel to make more accurate comparisons.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the scope of the present invention; any person who is familiar with this technique, without departing from the spirit and scope of the present invention, makes equal changes and modifications, all It should be covered within the patent scope of the present invention.

To sum up, the invention has the patent requirements of "industrial applicability", "novelty" and "progressiveness"; the applicant should file an application for an invention patent with the Jun Bureau in accordance with the provisions of the Patent Law.

10: Auxiliary system for underwriting claims 101: Processor 102:Database 103: Second Database 104: Keyword Extraction Subsystem 105:Keyword Parsing Subsystem 106: Data Marking Subsystem 107: Information Integration Subsystem 108: Front-end subsystem 109: Keyword Correction Subsystem 20: Information Devices ICD_T: Disease classification reference table MD_T: Medical Vocabulary Reference List IC: Insurance Clause MR: Electronic Medical Record SMR: Smart Medical Records SMR_T: Auxiliary interpretation reference table H1: First identification mark H2: Second identification mark COM: Annotation information L1: index link V: Viewer S: Searcher S_T: Specify search results F: Filter F_LIST: Illness inspection list F_ITEM: Filter items 30: Internet S: Implementation Method of Underwriting Claims Auxiliary System S10: Keyword Extraction S15: Keyword Correction S20: Grammar and Semantic Analysis S30: Comparison reference table S35: Provide search function S40: Compare insurance terms S50: Keyword Tag S60: Integrate intelligent medical records S70: Provide intelligent medical record viewing function S80: Provides filter function

FIG. 1 is a system architecture diagram of the present invention. Fig. 2 is a flow chart of the system implementation of the present invention. FIG. 3 is a schematic diagram (1) of the implementation of the present invention. FIG. 4 is a schematic diagram (2) of the implementation of the present invention. Fig. 5 is a schematic diagram (3) of the implementation of the present invention. Fig. 6 is a schematic diagram (4) of the implementation of the present invention. FIG. 7 is a schematic diagram (5) of the implementation of the present invention. Fig. 8 is another preferred embodiment (1) of the present invention. Fig. 9 is another preferred embodiment (2) of the present invention. Fig. 10 is another preferred embodiment (3) of the present invention.

S: Implementation Method of Underwriting Claims Auxiliary System

S10: Keyword Extraction

S20: Grammar and Semantic Analysis

S30: Comparison reference table

S40: Compare insurance terms

S50: Keyword Tag

S60: Integrate intelligent medical records

S70: Provide intelligent medical record viewing function

Claims (8)

An auxiliary system for underwriting and claim settlement, which can generate an intelligent medical record based on an electronic medical record for an information device to browse the intelligent medical record, comprising: a processor, another database, a second database, and a keyword extraction The subsystem, a keyword parsing subsystem, a data marking subsystem, an information integration subsystem and a foreground subsystem are respectively connected with the processor information; the database stores a disease classification reference table, a medical vocabulary reference table, an insurance clause and the electronic medical record; the keyword extraction subsystem is triggered by the processor to extract a plurality of keyword strings from the electronic medical record stored in the database, and the keyword strings include a medical keyword , a time string, a numerical string, and a symbol character; the keyword parsing subsystem is triggered by the processor to extract the results of the subsystem based on the keyword, so as to construct the grammatical structure and analysis of the electronic medical record A medical record time, and analyzing a semantic meaning of the keyword strings, and then comparing a plurality of medical terms from the medical vocabulary reference table based on the semantic meaning, and then referring to the disease classification based on the semantic meaning and the medical vocabulary The table parses out at least one disease standard code, and then judges whether a disease name corresponding to the disease standard code matches a disease keyword string of the insurance clause to generate a clause comparison result; the data marking subsystem is used for the processing The trigger is triggered to assign a plurality of visible first identification marks associated with the disease name to the keyword strings in the electronic medical record, and also to assign a reference table to an auxiliary interpretation reference table corresponding to the comparison result of the item. a second identification mark; The information integration subsystem is triggered by the processor to generate the auxiliary interpretation reference table stored in the second database, and then the electronic medical record, the first identification marks, the second identification mark and the auxiliary interpretation reference table , integrated into the intelligent medical record, wherein the auxiliary interpretation reference table at least includes the parsed disease name, the second identification mark and the disease standard code; the information device is connected to the front-end subsystem through a network to After reading the second database, the front-end subsystem provides a viewer for the information device, so that the intelligent medical record including the auxiliary interpretation reference table is displayed on the viewer; and the front-end subsystem also provides a coupling A searcher connected to the viewer, for the information device to define a specified search condition in the viewer, the specified search condition being the specific medical term that the keyword parsing subsystem wants to compare from the medical term reference table, Or the specific disease standard code to be parsed from the disease classification reference table, or the screening basis for judging whether the disease name matches the specific disease keyword string of the insurance clause. If applying for the underwriting and claims assistance system in the first item of the patent scope, wherein, the insurance clause is one of a health notification clause, an underwriting clause and a claim settlement clause or a combination thereof. For example, the underwriting and claim settlement assistance system of item 1 of the scope of the application further includes a keyword correction subsystem connected with the processor information for triggering the processor to retrieve the keywords extracted by the keyword extraction subsystem. Keyword strings, perform a spell check to compare multiple key phrases stored in the database extract the closest keyword sentence, and remove the meaningless or wrong part of the keyword string to generate the grammatical structure, the medical record time and the semantics for the keyword parsing subsystem to construct the electronic medical record The desired string of correction results. As claimed in claim 1 of the scope of the application, wherein the searcher can also be used by the information device to define a medical record time interval in the viewer, and the medical record time interval is the keyword parsing subsystem based on the keyword The extraction subsystem parses out the screening basis of the medical record time based on the retrieval results of the time strings. As claimed in claim 1 of the scope of the application, wherein the front-end subsystem is also provided for providing a filter coupled to the viewer, the filter including a disease inspection list, which includes one or more Screening items. Each screening item corresponds to each disease name parsed by the keyword analysis subsystem, and each screening item includes an index link that can lead to the keyword string in the electronic medical record. An implementation method of an underwriting and claims assistance system, comprising: a keyword extraction step: a keyword extraction subsystem extracts multiple keyword strings from an electronic medical record, and the keyword strings include a medical keyword, a time string, a numeric string, and a symbolic character; A grammatical and semantic analysis step: a keyword parsing subsystem constructs a grammatical structure of the electronic medical record, analyzes a medical record time, and analyzes a semantics of the keyword strings based on the retrieval results of the keyword extraction subsystem; a comparison The step of comparing the reference table: the keyword parsing subsystem compares a plurality of medical terms from a medical vocabulary reference table based on the semantics, and then parses at least one disease classification reference table based on the semantics and the medical vocabulary. Disease standard code and a disease name corresponding to the disease standard code; a comparison insurance clause step: the keyword parsing subsystem determines whether the disease name corresponding to the disease standard code matches a disease keyword string of an insurance clause , to generate an item comparison result; a keyword marking step: a data marking subsystem assigns a plurality of visible first identification marks associated with the disease name to the keyword strings in the electronic medical record, and also marks the keyword strings in the electronic medical record. An auxiliary interpretation reference table is assigned a second identification mark corresponding to the comparison result of the item; a step of integrating intelligent medical records: an information integration subsystem generates the auxiliary interpretation reference table, and then the electronic medical record, the first identification The mark, the second identification mark and the auxiliary interpretation reference table are integrated into an intelligent medical record, wherein the auxiliary interpretation reference table at least includes the parsed disease name, the second identification mark and the disease standard code; a Steps of providing an intelligent medical record viewing function: a front-end subsystem provides the information device with a viewer, so that the intelligent medical record including the auxiliary interpretation reference table is displayed on the viewer; and a step of providing a searcher function: the front-end screen The system provides a searcher for the information device for the information device to define a specified search query in the viewer disease, a medical record time interval, wherein the specified search disease is the specific medical term to be compared by the keyword parsing subsystem, the specific disease standard code to be parsed, or the specific disease keyword string to be judged whether it matches filter basis. For example, the implementation method of the underwriting and claim settlement assistance system according to item 6 of the scope of the patent application further includes a keyword correction step: a keyword correction subsystem executes the keyword strings extracted after the keyword extraction step, and executes a spelling check, to compare the closest key phrase from the multiple key phrases stored in the database, and remove the meaningless or wrong part of the key string to generate the When the step of comparing the reference table is executed, the grammatical structure of the electronic medical record is constructed, the time of analyzing the medical record, and the correction result of a keyword string required for the semantics. For example, the implementation method of the underwriting and claim settlement assistance system in the sixth item of the patent application scope further includes a step of providing a filter function: the front-end subsystem provides a filter for the information device, and the filter includes a disease inspection list, which includes One or more screening items, each of which corresponds to each of the disease names parsed during the execution of the comparison reference table step, and each of the screening items includes the keyword that can lead to the electronic medical record An indexed concatenation of strings.

Images