Inpatient-oriented hospitalization medical behavior acquisition, analysis and supervision system

Abstract

The invention relates to an inpatient-oriented hospitalization medical behavior acquisition, analysis and supervision system. The system comprises an admission recording module, a disease course recording module, a nursing recording module and a doctor advice management module; based on voice recognition, natural language processing and other technologies, a structured medical record document is automatically generated, and manual errors and repeated labor are reduced; the illness state of a patient can be evaluated in real time, disease course records are automatically generated, and critical value automatic early warning is achieved; the nursing recording module monitors vital signs through an early warning scoring engine, automatically calculates early warning total time division and synchronously and actively pushes alarm information to a doctor on duty or a doctor in bed management, a traditional one-way communication mode of'instruction-response 'is broken through, information is ensured to be direct, and omission is avoided; the doctor's advice management module supports voice issuing and intelligent auditing; the system can also automatically generate a whole-process record covering department transfer, shift change and hospital discharge, and seamless transmission of medical information is guaranteed; through the linkage of the modules, the medical service quality, safety and digitization level are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical information management, in particular to a hospitalization medical behavior collection, analysis and supervision system for inpatients. BACKGROUND

[0002] With the continuous advancement of hospital informationization construction, electronic medical record systems, nursing information systems, examination and inspection systems, and medical order management systems have been put into use one after another. Clinical data collection and medical record document management are gradually transforming from paper-based to digital. In most hospitals, doctors and nurses can complete admission records, medical records, nursing records, medical order issuance, and examination and inspection result queries in the information system. Some hospitals have also introduced mobile ward round systems, voice input modules, and bedside monitoring device networking functions, which have improved the efficiency and availability of information input. Overall, inpatient management is developing towards standardization and digitization.

[0003] However, the current hospital-wide "electronic medical record digitization" still mainly focuses on document electronicization and storage electronicization. The system mainly serves as a "document input tool" and an "information display window". When writing admission records, first medical records, daily medical records, and transfer records, doctors still need to manually organize a large amount of consultation information, examination and inspection reports, and nursing record content. Nursing staff also need to repeatedly input a large number of structured fields when recording vital signs, executing medical orders, observing the patient's condition, and handing over shifts. Such high-frequency manual operations not only consume time and are prone to errors, but also cause medical staff to spend a lot of time on document work, leaving them less time for diagnosis and treatment and nursing.

[0004] In the prior art, although some electronic medical record systems provide voice input functions, template application functions, or limited intelligent auxiliary modules, they still cannot solve the following key technical problems in the hospitalization process:

[0005] Firstly, clinical information is scattered and data structure is not unified. Admission consultation data, medical record data, nursing data, medical order data, and monitoring device data are stored in different systems or sub-modules. There is a lack of automatic fusion and semantic association analysis mechanism across systems, which leads doctors to manually review multiple records to obtain the complete path of the patient's condition evolution.

[0006] Secondly, the identification and processing link of critical values still highly depends on manual judgment. Most current systems can only pop up a prompt when a critical value appears, then call the nurse, the nurse confirms, informs the doctor to sign and review, and gives a processing suggestion. The entire process cannot automatically generate a medical record containing the processing suggestion, which easily causes a delay between "discovery - record - medical order - execution", increasing the clinical risk.

[0007] The nursing record lacks deep linkage with the progress record. The vital signs, nursing measures, assessment scores and the like in the nursing record cannot be automatically linked with the progress record, and doctors have difficulty in obtaining real-time nursing observation results in time, and nursing risk scores such as MEWS cannot be automatically written into the progress document or provided to the system for cross-module analysis.

[0008] The doctor's order review is still mainly based on rule matching, and the intelligence is insufficient. The existing doctor's order review system is usually based on a static knowledge base or rule base, and cannot make dynamic judgments according to the changes in the patient's condition, the medication link, the progress record and the like, and cannot make inferences on the associated path between new doctor's orders and the patient's state based on a medical knowledge graph, so it is difficult to accurately identify individualized conflict risks.

[0009] In addition, the handover record and the transfer record rely on manual sorting. At present, nurses often need to manually summarize abnormal vital signs, execution and uncompleted doctor's orders of the previous shift; doctors need to reorganize the admission situation, previous treatment history, multiple inspection results and nursing assessment data for the transfer record, which not only has a large workload, but also is prone to problems such as information omission, inconsistent content or inconsistency with the actual execution situation.

[0010] The reason why the existing technology cannot solve the above problems lies in:

[0011] 1) Lack of unified data structure and multi-source fusion mechanism, unable to automatically integrate heterogeneous data from different systems in real time;

[0012] 2) Lack of technical framework combining natural language processing, multi-modal data analysis and medical knowledge graph reasoning together;

[0013] 3) Lack of event-driven mechanism for cross-module linkage, unable to realize the whole-process automation closed loop from abnormality discovery to document generation, doctor's order pushing, nursing completion and record signature;

[0014] 4) Lack of intelligent document system capable of automatically generating structured progress records, nursing records, handover records and transfer records, resulting in that the inpatient management is still in a state of "digital inefficiency".

[0015] Therefore, it is urgent to develop an inpatient medical behavior acquisition, analysis and supervision system for inpatients to solve the above problems. SUMMARY

[0016] The purpose of the present application is to solve the technical problems raised in the background art, and to provide an inpatient medical behavior acquisition, analysis and supervision system for inpatients, comprising:

[0017] an admission record module, a progress record module, a nursing record module and a doctor's order management module;

[0018] The admission record module is configured to acquire the doctor's inquiry information, and generate a structured admission record according to medical standards and a preset template.

[0019] The disease course record module is configured to integrate multi-source medical data and generate a structured disease course record. When a critical value is identified, the module triggers an early warning process and generates a disease course record containing recommended treatment orders. The disease course record generated by the disease course record module includes at least a first disease course record, a ward round record, and a discharge record.

[0020] The nursing record module is configured to record the entire nursing process. The vital signs and assessment data collected by the module are visible to the disease course record module, and the module has a built-in disease monitoring unit configured to start a hierarchical early warning and reporting process according to preset rules when an abnormality is detected.

[0021] The medical order management module is configured to receive medical orders input in the form of voice or instructions, convert the medical orders into structured data, perform intelligent auditing, and generate a refill or return medication prompt based on the use of the medication.

[0022] The admission record module, the disease course record module, the nursing record module, and the medical order management module realize data intercommunication and jointly constitute a medical closed-loop management architecture from information collection, abnormality identification, decision generation, to execution feedback.

[0023] As a preferred technical solution of the present application, the system further comprises a doctor-patient communication record module and an out-of-hospital follow-up module.

[0024] The doctor-patient communication record module is configured to record the entire doctor-patient communication process and automatically generate patient-readable disease description and health guidance materials.

[0025] The out-of-hospital follow-up module is configured to automatically develop a personalized follow-up plan based on the patient's disease characteristics.

[0026] As a preferred technical solution of the present application, the admission record module comprises an adaptive learning unit configured to learn the diagnosis and treatment habits and writing styles of different doctors and make individualized adjustments when generating records.

[0027] As a preferred technical solution of the present application, the disease course record module supports voice recognition of rescue instructions and automatically generates rescue records. The various disease course records generated by the module are synchronized in real time to the nursing record module and are visible to all medical staff.

[0028] As a preferred technical solution of the present application, the records generated by the disease course record module further include blood transfusion records, critical value processing records, operation records, surgery records, death records, transfer-out records, transfer-in records, shift records, handover records, stage summaries, and medical certification documents.

[0029] As a preferred technical solution of the application, the illness monitoring unit of the nursing record module is internally provided with an early warning scoring engine, which is used to calculate a total early warning score according to the scoring results of the patient's heart rate, systolic pressure, respiratory rate, body temperature and consciousness level, and compare the total early warning score with a preset threshold to determine an early warning level and start a corresponding reporting process.

[0030] As a preferred technical solution of the application, the medical order management module is internally provided with a conflict detector based on a medical knowledge graph, which is used to traverse the associated path of a new medical order and the current state of a patient in the knowledge graph to identify potential risks.

[0031] As a preferred technical solution of the application, the medical order management module further includes a medical insurance pre-audit unit, which is used to pre-audit the medical order based on medical insurance rules before the medical order is audited, so as to reduce the violation of medical orders or non-reimbursable items.

[0032] As a preferred technical solution of the application, the nursing record module includes a signature and audit unit, which is used to support nursing staff to automatically complete electronic signature and record audit through voice or instructions.

[0033] The application further provides an electronic device, which includes a memory, a processor and a computer program stored in the memory and executable on the processor, and when the processor executes the program, the control and data processing logic of the system are realized.

[0034] Compared with the prior art, the application has the following beneficial effects:

[0035] Through the collaborative work of the admission record module, the medical history record module, the nursing record module and the medical order automatic management module, the application realizes the intelligentization of medical staff's document processing; the admission record module automatically extracts key information from the doctor's inquiry voice based on voice recognition technology and natural language processing (NLP) to generate structured admission records; this reduces the workload of doctors manually entering medical records and avoids writing errors caused by fatigue or negligence; in addition, the medical history record module integrates multi-source medical data (including laboratory data, image data and nursing records, etc.), and through the automatic illness severity scoring and critical value early warning functions, it helps doctors efficiently and accurately grasp the changes of the patient's condition; these innovative technologies greatly save the time of medical personnel and improve the work efficiency, and reduce the medical risks caused by manual errors;

[0036] The disease course record module combines medical knowledge graph and adaptive learning function, can intelligently evaluate the patient's condition and generate disease course record in real time, especially has obvious advantages in critical value identification; when the test results returned by the laboratory or bedside monitoring equipment exceed the preset critical value threshold, the system automatically triggers the early warning mechanism, and pushes the processing suggestions (such as drug adjustment, treatment means) to the medical order module, to ensure the timely update and execution of the treatment plan; the nursing record module monitors the patient's vital signs in real time through the built-in early warning score engine (such as MEWS score), and calculates the early warning total score; when the early warning total score reaches the high-risk threshold, the system automatically reminds the doctor and nursing staff to intervene; through this technical scheme, the system can timely identify the condition change and automatically generate the corresponding emergency treatment plan, greatly reducing the risk of clinical decision delay and medical errors;

[0037] The close integration of the medical order automatic management module and the nursing record module ensures efficient cooperation and information sharing of medical staff when executing medical orders; the medical order automatic management module can support voice input, and convert voice into structured data for intelligent audit, avoiding omissions and errors of manual audit; during the execution of medical orders, the system automatically records the execution and ensures that each medical order is executed in time and accurately through cross-module linkage; the system not only continuously tracks the condition during hospitalization, but also during transfer and shift handover, through automatically generating transfer records and shift handover records, to ensure that information is not missed or incorrect; when the transfer record is generated, the system automatically integrates the patient's admission diagnosis, disease course record, treatment history, medical order execution, nursing record and other information, avoiding information omissions, errors or inconsistencies that may occur during manual sorting and recording. BRIEF DESCRIPTION OF DRAWINGS

[0038] Fig. 1 is the system block diagram of the inpatient medical behavior collection, analysis and supervision system for inpatients proposed in the embodiments of the present application;

[0039] Fig. 2 is the medical order audit and risk control flowchart of the inpatient medical behavior collection, analysis and supervision system for inpatients proposed in the embodiments of the present application. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical scheme and advantages of the present application more clear, the following combines embodiments and drawings to further describe the present application. Figs. 1-2 The present application is further described in detail. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0041] The following describes in detail the structure, data processing path, algorithm derivation chain, and collaborative mechanism of a system for collecting, analyzing, and monitoring inpatient medical behavior, based on a typical process of the present invention in a hospital inpatient setting. This embodiment focuses on "from information collection to early warning and conflict identification, and finally to the formation of a unified risk control indicator", clearly demonstrating the overall logic of the present invention to those skilled in the art.

[0042] The system of the present invention consists of an admission record module, a medical record module, a nursing record module, and a medical order management module as its basic framework, and is combined with a doctor-patient communication record module, an outpatient follow-up module, an adaptive learning unit, a rescue record unit, an early warning scoring engine, a conflict detector based on medical knowledge graph, a medical insurance pre-approval unit, and an electronic signature and review unit to form a complete closed loop.

[0043] In the hospital admission scenario, the patient receives a doctor's consultation, and the doctor continuously describes the patient's medical history via voice. The admission record module first transcribes the voice into a text sequence T = {w1, w2, ..., w...} n}, where T represents the sequence of consultation texts; w i This represents the i-th word in the text; n represents the total number of lexical units. The text sequence is encoded using a medical language model to obtain a semantic vector.

[0044] E = f(T);

[0045] Where E represents the encoded semantic vector; f(·) is the text encoding function.

[0046] The system pre-sets a set of specialized subject template vectors {U1, U2, ..., U...} K The relationship between the text and the template is calculated using vector cosine similarity.

[0047]

[0048] Where, sim j U represents the similarity between the j-th template and the current consultation. j Here, "·" represents the template vector; "·" represents the inner product; and "∥·∥" represents the vector magnitude. The system selects the template with the highest similarity as the structure of the current patient's admission document, i.e.:

[0049]

[0050] Where, j * Indicates the selected best template index; argmax j This represents the template number that maximizes the similarity.

[0051] Subsequently, the adaptive learning unit adjusts the doctor's personalized style based on the template. Let R be the initial record vector generated by the template, and P be the doctor's personal style vector.d , the department standard style is P0, and the corrected record is recorded as:

[0052] R' = R + λ(P d -P0);

[0053] wherein, R ' is the personalized adjusted admission record vector; λ is a style fusion coefficient for controlling the balance between personalization and standardization; P d is the doctor's habit feature vector; P0 is the department standard vector, and the record enters the system as the initial state vector of the subsequent course record module.

[0054] After the above steps, the admission record generated by the admission record module not only conforms to the medical standard and template structure, but also gradually fits the writing habits of different doctors, and finally stores in the patient main index of the system in the form of structured data, providing a unified data starting point for the subsequent course record generation and follow-up record generation.

[0055] During hospitalization, the course record module continuously collects multi-source data from HIS, LIS, PACS, nursing record module and monitoring equipment, and represents the overall state of the patient at a certain time as a feature vector:

[0056] Z = [z1, z2, …, z m ];

[0057] wherein, Z represents the multi-source comprehensive feature; z k is the kth feature, such as body temperature, heart rate, test index, etc.; m is the number of features.

[0058] Since the dimensions and value ranges of different features are different, the course record module first normalizes each feature to obtain a normalized feature vector:

[0059]

[0060] wherein, z ' k is the normalized value of the kth feature; μ k is the mean of the kth feature in the historical data; σ k is the standard deviation of the kth feature in the historical data. After standardization, all features are mapped to a unified scale, which is conducive to subsequent risk scoring and early warning calculation.

[0061] On this basis, the course record module assigns weights to different features according to their importance in disease progression to form a disease severity score:

[0062]

[0063] wherein, Rmed R is the severity score of the disease; a k is the weight of the kth feature. The weight value is determined according to clinical experience and statistical learning results; z k indicates the kth normalized feature value described above; m is the total number of features. The score gives a continuous scale of the severity of the disease in numerical value, and the current disease risk level description can be automatically brought out in the subsequent course record.

[0064] When the system receives the test results from the laboratory information system, the course record module compares the relevant indicators with the pre-set critical value threshold. For a test item, its result is denoted as v, and its critical value threshold is denoted as θ, then the critical event trigger identifier is:

[0065]

[0066] wherein E critical indicates whether a critical event is triggered, v indicates the actual measurement value of the test; θ indicates the critical value threshold of the test item; "1" indicates that the critical value is triggered, and "0" indicates that the critical value is not triggered.

[0067] When E critical = 1, the course record module automatically generates a course record entry containing the critical value description, and according to the specific test item and the current disease severity score R med retrieves the corresponding recommended treatment order set from the built-in knowledge base, such as fluid replacement, potassium supplementation, emergency imaging examination, etc. After the doctor confirms the recommended set on the mobile terminal interface, the selected treatment order is directly transmitted to the nurses' station through the order management module, realizing a second-level response closed loop from test results to treatment execution. If a critical event is triggered, the corresponding course record is automatically generated and the treatment scheme library is called to recommend candidate orders for the doctor. After the doctor confirms, the order enters the order management module.

[0068] At the same time, the nursing record module continuously collects the vital signs and assessment data of the patient throughout the hospitalization process, forming time series data. In order to realize early warning, the nursing record module is built-in early warning scoring engine, which scores the key parameters such as heart rate, respiratory rate, systolic blood pressure, body temperature and consciousness level respectively. Let the single score corresponding to different physiological parameters be s hr ,s rr ,s bp ,s temp ,s avpu , then the comprehensive warning score is:

[0069] S total = s hr + s rr + s bp + s temp + savpu ;

[0070] wherein, S total represents the total score of early warning; s hr represents the score value corresponding to heart rate; s rr represents the score value corresponding to respiratory rate; s bp represents the score value corresponding to systolic blood pressure; s temp represents the score value corresponding to body temperature; s avpu represents the score value corresponding to the level of consciousness (awake / answer when called / pain stimulus response / no response, etc.).

[0071] The system presets two early warning thresholds θ1, θ2, and the system automatically divides the early warning levels:

[0072]

[0073] wherein, L warn represents the current illness early warning level; L3 represents high-level early warning; L2 represents medium-level early warning; L1 represents low-level monitoring state; S total is the above early warning total score; θ1 and θ2 are medium and high-level early warning thresholds, respectively.

[0074] In order to unify the illness severity score of the medical record module and the early warning total score of the nursing record module into a comprehensive illness risk perspective, the system combines the two to form a comprehensive illness risk index. The system combines the illness severity score R med of the medical record module and the nursing early warning score S total as follows:

[0075] R global = β1R med + β2S total ;

[0076] wherein, R global represents the comprehensive illness risk index; R med represents the illness severity score calculated by the medical record module; S total represents the early warning total score obtained by the nursing record module; β1 and β2 represent the weight coefficients of the two in the comprehensive risk, which can be adjusted according to different disease types or department characteristics.

[0077] When the doctor issues a new drug order for the patient, the order management module receives the order content input in the form of voice or instruction, and after voice recognition and semantic analysis, the order is analyzed into a drug entity M and structured fields such as dosage, frequency, route of administration, etc. Subsequently, the conflict detector based on the medical knowledge graph traverses the drug entity M and the current patient state entity set M and the patient state entity set P = {p1, p2, …, p kthe associated path in the knowledge graph, and calculate the risk score:

[0078]

[0079] wherein Risk(M) represents the conflict risk score of the current new medical order; M represents the drug entity corresponding to the newly issued medical order; P = {p1, p2, …, pi, …, pP} represents the set of patient-related state entities, including past allergies, underlying diseases, current medication, etc.; Path(M, pi) represents the associated path between the drug entity M and the i-th patient state entity pi in the knowledge graph; g(·) represents a scoring function for risk assessment of an associated path; wi represents the weight coefficient of the i-th patient factor in the conflict risk; and k represents the number of patient state entities. k i i i

[0080] The knowledge graph conflict detector compares the medical order conflict risk value with a preset threshold value δ:

[0081]

[0082] wherein E drug represents a flag indicating whether the medical order conflict event is triggered; Risk(M) is the medical order conflict risk score described above; and δ represents the conflict risk threshold value, wherein “1” indicates that there is a serious conflict that needs to be intercepted, and “0” indicates that no serious conflict is detected.

[0083] In order to evaluate the overall risk of the patient from the management level, the system further combines the comprehensive disease risk index and the medical order conflict risk to construct a total risk control index:

[0084] G = γ1R global + γ2Risk(M);

[0085] wherein G represents the total risk control index, which is used to comprehensively reflect the current disease progression and medication safety risk; R global represents the comprehensive disease risk index described above; Risk(M) represents the conflict risk score of the new medical order; and γ1 and γ2 represent the weight parameters of the disease risk and the medication risk in the total risk assessment.

[0086] The system can automatically trigger different levels of management strategies according to different value ranges of G, for example, when the total risk index exceeds a certain threshold value, the superior doctor is forced to review the medical record and medical order, or the case is automatically pushed to the quality control personnel for special examination. The medical insurance pre-audit unit is embedded in the medical order review link, and the structured medical order field is matched with the medical insurance rule library. When the medical order item belongs to the limited reimbursement or non-reimbursable item, a prompt is given to the doctor so that he can make necessary adjustments. ​​​​

[0087] In the nursing record module, the signature and audit unit allows nursing staff to complete infusion, dressing change, special operation or high-risk drug, through voice password or instruction triggered electronic signature process. The system binds the signature event with the corresponding operation record and timestamp, and writes it into the audit log chain, so as to realize the traceability and irrefutability of nursing behavior.

[0088] The doctor-patient communication record module records the key communication process before and after the patient is hospitalized and discharged. For example, the important communication content with the patient and the family about the treatment plan, prognosis evaluation, precautions, etc. can be converted into patient understandable text explanation through audio recording and summary generation algorithm. The module uses a similar text coding and template matching mechanism as the admission record to automatically replace professional terms with popular explanations and generate patient readable materials.

[0089] The off-site follow-up module reads the final diagnosis information, main treatment plan and medication plan in the disease record module when the patient is discharged, matches it with the standard follow-up path library, forms a personalized follow-up plan, including follow-up time points, symptoms to pay attention to, examination items to review, etc., and can be pushed to the patient through a mobile application.

[0090] The present application also provides a corresponding electronic device implementation form, which includes a memory, a processor and a computer program stored on the memory and executable on the processor. When the processor executes the program, it realizes the control logic and data processing process of the above-mentioned admission record module, disease record module, nursing record module, medical order management module, doctor-patient communication record module, off-site follow-up module, adaptive learning unit, early warning scoring engine, conflict detector based on medical knowledge graph, medical insurance pre-audit unit and signature and audit unit. Through the above-mentioned hardware and software cooperation, the system of the present application can be deployed and run in the actual hospital environment, and complete the automatic, intelligent collection, analysis and supervision of inpatient medical behavior.

[0091] The technical solution of the present application will be further described below in conjunction with an actual case.

[0092] Patient Zhang Guoguo, male, 58 years old, admitted to the department of cardiology for "chest pain for 3 hours". The admitting doctor Dr. Wang directly inquired at the bedside in a voice manner, and the system intelligently recognized the entire inquiry process and automatically arranged it into the inquiry illness information through the admission record module: "The patient has a squeezing pain in the chest for 3 hours, radiating to the left shoulder and back, accompanied by heavy sweating, no coughing, no hemoptysis, no obvious inducement, and a history of hypertension for 10 years."

[0093] The admission record module first converts the voice into a text sequence T={w1,w2,…,w n} and then gets semantic vector E = f(T) through encoding function f(·). The system presets multiple admission templates of the cardiology department, denoted as vector set {U1, U2, U3}, which respectively correspond to templates such as “chest pain suspected coronary heart disease”, “admission of heart failure”, “admission of arrhythmia”, etc. The system calculates the matching degree of the current inquiry and each template through cosine similarity, for example:

[0094] sin1 = 0.92, sin2 = 0.61, sin3 = 0.45;

[0095] wherein, represents the similarity of the inquiry semantic vector and the jth template vector, and the maximum value in this example is sin1 = 0.92. The system automatically selects template 1, i.e. the “chest pain suspected coronary heart disease” template, as the structural basis of the admission record. In the formula, E represents the semantic vector of the encoded inquiry text; U j represents the semantic vector of the jth candidate template; ||·|| represents the vector length; and sin j represents the similarity with the jth template. After the template is selected, the self-adaptive learning unit makes fine adjustments to the content of the medical record according to the record style of Dr. Wang in the past.

[0096] Suppose that the personal style vector of Dr. Wang recorded in the system is P d , the standard style vector of the department is P0, and the initial record vector generated according to the template is R. The system updates according to R' = R + λ(P d -P0). Suppose that λ = 0.4, and (P d -P0) = 0.5 in a certain vector dimension, then the adjustment amount of this dimension is 0.4 × 0.5 = 0.2, so that the final record R ' is closer to the habitual language of Dr. Wang. In the formula, R represents the initial record vector generated according to the template; R ' represents the record vector after style adjustment; P d represents the personal style vector of the current doctor; P0 represents the standard style vector of the department; and λ represents the style adjustment weight.

[0097] After admission, the system obtains the multi-source data of the patient from HIS, LIS, PACS, nursing record module and bedside monitoring equipment to form feature vector Z. At a certain time, the collected key data is as follows: body temperature 36.8℃, heart rate 118 beats / min, respiratory rate 26 times / min, systolic pressure 95 mmHg, and potassium 2.8 mmol / L. The system statistically obtains the mean μ hr = 80 and standard deviation σ hr = 10 of the heart rate, and the mean μ K = 4.0 and standard deviation σ K = 0.5 of the potassium. The standardization is calculated as follows:

[0098]

[0099] where: z ' hr z represents the normalized value of heart rate; z ' K z represents the normalized value of potassium; 118 and 2.8 are the current measured values; 80 and 4.0 are the historical mean values; 10 and 0.5 are the corresponding standard deviations.

[0100] The system assigns weight coefficients to different features, for example, heart rate weight α hr = 0.25, blood pressure weight α bp = 0.20, respiratory rate weight α rr = 0.15, potassium weight α K = 0.30, temperature weight α temp = 0.10, assuming that after normalization, the normalized values corresponding to the five features are: heart rate z ' hr = 3.8, blood pressure z ' bp = -1.0, respiration z ' rr = 1.6, potassium z ' K = -2.4, temperature z ' temp = 0.2, the disease progression record module calculates the severity score:

[0101] R med = 0.25 x 3.8 + 0.20 x (-1.0) + 0.15 x 1.6 + 0.30 x (-2.4) + 0.10 x 0.2;

[0102] R med = 0.95 - 0.20 + 0.24 - 0.72 + 0.02 = 0.29;

[0103] where: R med represents the severity score; each coefficient α represents the importance of the corresponding feature in the disease assessment; each z ' is the normalized feature value, although hypokalemia is significantly abnormal, due to the weight configuration and the comprehensive effect of other indicators, the current overall score is 0.29; if the subsequent indicators worsen, the value will rise accordingly.

[0104] The system receives the potassium test result v = 2.8 mmol / L from the LIS and compares it with the preset critical value threshold θ = 3.0, and the critical value triggering rule is:

[0105]

[0106] In this example, θlow = 3.0, since 2.8 < 3.0, so E critical = 1, the system considers that a low potassium critical event is triggered, automatically generates a "critical value record" and a draft of a progress note containing key data, where: E critical is a critical event trigger mark; v is the value of this test; θ low is the critical low value threshold of the corresponding item.

[0107] At this time, the nursing record module simultaneously acquires the patient's current vital signs from the monitor and automatically calculates the early warning score, assuming that the scoring rules are as follows: 0 points for heart rate 51-100, 1 point for 101-110, and 2 points for > 110; 0 points for respiration 9-20, 1 point for 21-24, and 2 points for 225; 0 points for systolic blood pressure 100-199, 1 point for 80-99, and 3 points for < 80; 0 points for body temperature 36-38℃, 1 point for < 36 or > 38; 0 points for consciousness, and 3 points for others. The heart rate of this example is 118 beats per minute, which is s hr = 2, the respiration is 26 times per minute, which is s rr = 2, the systolic blood pressure is 95 mmHg, which is s bp = 1, the body temperature is 36.8℃, which is s temp = 0, the consciousness is s avpu = 0.

[0108] The system calculates according to total = s hr + s rr + s bp + s temp + s avpu = 2+2+1+0+0 = 5, where: S total is the total score of the warning; s hr is the single item score of heart rate; s rr is the single item score of respiratory rate; s bp is the single item score of systolic blood pressure; s temp is the single item score of body temperature; s avpu is the single item score of consciousness level, the system sets the medium warning threshold θ1 = 3 and the high warning threshold θ2 = 5, according to

[0109]

[0110] In this example, S total = 5, which is exactly equal to θ2, so L warn = L3, that is, a high-level warning, the system automatically pushes a high-priority alarm to the current on-duty doctor and the responsible nurse, and forces the nurse to complete a nursing record, recording "the patient's heart rate and respiration have increased, and has been treated according to the doctor's orders" and other content. In order to unify the progress score and nursing warning on the whole level, the system calculates the comprehensive illness risk index:

[0111] R global = β1R med + β2S total ;

[0112] If β1= 0.5, β2= 0.5, then:

[0113] R global = 0.5x0.29 + 0.5x5 = 0.145 + 2.5 = 2.645;

[0114] wherein: R global is the comprehensive risk index of the illness; β1, β2 are the weights of the illness severity and the nursing warning; R med is the illness score output by the illness record module; S total is the total warning score output by the nursing record module.

[0115] The index value is written into the illness record and the shift handover record as a quantitative index for the attention of the on-duty nurses and doctors. Subsequently, Dr. Wang decides to issue the medical order of potassium supplement and adjustment of the antihypertensive drug for the patient, and issues the instructions such as "10% potassium chloride 10ml is added to 0.9% sodium chloride 250ml for intravenous infusion, once a day" and "the dose of the antihypertensive drug is adjusted" through voice.

[0116] The medical order automation management module parses the voice into a structured medical order object M, and reads the patient state entity set P = {p1, p2} at the same time, wherein p1 is "previous hypertension" and p2 is "current hypokalemia". The medical knowledge graph gives a low risk of Path(M, p1) and a moderate risk of Path(M, p2) (for example, "fast potassium supplement may cause arrhythmia, and ECG needs to be closely monitored"). Assuming that the system gives a risk score of g(Path(M, p1)) = 0.2 and g(Path(M, p2)) = 0.7 for the two paths, and the weights are w1 = 0.3 and w2 = 0.7 respectively, then:

[0117] Risk(M) = 0.3x0.2 + 0.7x0.7 = 0.06 + 0.49 = 0.55;

[0118] wherein: Risk(M) is the conflict risk value of the new medical order; g(Path(M, p i )) is the risk score of the drug-patient state path in the knowledge graph; w i is the weight of different factors in risk assessment, and the risk threshold δ = 0.8 in this example. Since Risk(M) = 0.55 < 0.8, the system judges that there are matters needing attention, but it does not belong to the serious contraindication that must be intercepted, so it increases the prompt of "close monitoring" on the medical order in the form of "high risk warning".

[0119] The system can also calculate a total risk control index according to the comprehensive condition risk index and the medical order risk, for example:

[0120] G = γ1R global + γ2Risk(M);

[0121] If γ1 = 0.7, γ2 = 0.3, and R global = 2.645, Risk(M) = 0.55, then we have:

[0122] G = 0.7 x 2.645 + 0.3 x 0.55 = 1.8515 + 0.165 = 2.0165;

[0123] In the formula, G is the total risk control index, which is used to provide a general quantitative reference for quality control personnel and managers; γ1 and γ2 are the weights of the condition risk and the medication risk in the total risk.

[0124] As the treatment progresses, the patient's condition gradually stabilizes, and the vital signs tend to be normal, and the early warning total score gradually decreases. The nursing record module automatically generates vital sign record sheets, medical order execution record sheets, nursing record sheets, and surgical nursing record sheets, etc. Before each shift handover, the system automatically generates a shift handover record draft based on the condition changes, early warning records, and uncompleted medical orders of the previous shift, which automatically brings out the R global and the key matters needing attention for the handover; the on-duty nurse checks and supplements the text description, completes the electronic signature, and forms a complete shift handover record.

[0125] Before transferring to another department, the transfer-out record is automatically generated and confirmed by the cardiologist through the system. The system fills in the transfer-out record content according to the preset template: including the admission date, transfer-out date, transfer-out department, transfer-in department, patient name, gender, age, chief complaint (extracted from the admission record module), admission condition (automatically integrated physical examination and laboratory examination results), admission diagnosis, diagnosis and treatment process (the system automatically sorts out "what checks were done", "what drugs were used", "when potassium was supplemented", "when coronary angiography was performed", etc. from the medical record and medical order execution record), current condition (call the latest condition assessment and nursing summary), current diagnosis, transfer purpose and matters needing attention (automatically bring out the key monitoring items and medication attention), and electronic physician signature field. The doctor only needs to modify and confirm a small amount of key description to complete the transfer-out record. At the same time, the nursing end also automatically generates a nursing summary at the time of transfer, which summarizes the vital sign trend, main nursing measures, assessment results, and matters needing attention for handover, which will be reflected in the "transfer handover record".

[0126] The above merely describes preferred embodiments of the present application, and is not used to limit the present application, any modification, equivalent replacement and improvement within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A system for collecting, analyzing, and monitoring inpatient medical behavior, characterized in that, include: The module includes admission record, progress note, nursing record, and medical order management. The admission record module is used to obtain the doctor's consultation information and generate a structured admission record based on medical standards and preset templates. The medical record module is used to integrate multi-source medical data and generate structured medical records; when a critical test value is identified, an early warning process is triggered and a medical record containing recommended treatment orders is generated; the medical record generated by the medical record module includes at least: the initial medical record, ward round record, and discharge record; The nursing record module is used to record the entire nursing process. The vital signs and assessment data collected by the module are visible to the medical record module. It also has a built-in disease monitoring unit, which is used to initiate a graded early warning and reporting process according to preset rules when an abnormality is detected. The medical order management module is used to receive medical orders input by voice or command, convert the medical orders into structured data and perform intelligent review, and generate medication renewal or return prompts based on medication usage. The admission record module, medical record module, nursing record module, and medical order management module achieve data interoperability and together form a closed-loop medical management architecture from information collection, anomaly identification, decision generation to execution feedback.

2. The inpatient medical behavior collection, analysis, and monitoring system for inpatients according to claim 1, characterized in that, The system also includes a doctor-patient communication record module and an outpatient follow-up module; The doctor-patient communication record module is used to record the entire doctor-patient communication process and automatically generate patient-readable descriptions of their condition and health guidance materials. The outpatient follow-up module is used to automatically develop personalized follow-up plans based on the patient's disease characteristics.

3. The inpatient medical behavior collection, analysis, and monitoring system for hospitalized patients according to claim 1, characterized in that, The admission record module includes an adaptive learning unit, which learns the diagnostic and treatment habits and writing styles of different doctors and makes personalized adjustments when generating records.

4. The inpatient medical behavior collection, analysis, and monitoring system for inpatients according to claim 1, characterized in that, The medical record module supports voice recognition of emergency commands and automatically generates emergency records. The various medical records generated by the module are synchronized to the nursing record module in real time and are visible to all medical staff.

5. A system for collecting, analyzing, and monitoring inpatient medical behavior according to claim 4, characterized in that, The records generated by the medical record module include, but are not limited to, blood transfusion records, critical value management records, operation records, surgical records, death records, transfer-out records, transfer-in records, shift handover records, shift handover records, periodic summaries, and medical certificates.

6. The inpatient medical behavior collection, analysis, and monitoring system for inpatients according to claim 1, characterized in that, The nursing record module's condition monitoring unit has a built-in early warning scoring engine, which is used to calculate the total warning score based on the patient's heart rate, systolic blood pressure, respiratory rate, body temperature, and level of consciousness. The total warning score is then compared with a preset threshold to determine the warning level and initiate the corresponding reporting process.

7. A system for collecting, analyzing, and monitoring inpatient medical behavior according to claim 1, characterized in that, The medical order management module has a built-in conflict detector based on medical knowledge graph, which is used to traverse the association path between new medical orders and the patient's current status in the knowledge graph to identify potential risks. The medical order management module also includes a medical insurance pre-review unit, which is used to pre-review the medical orders based on medical insurance rules before the medical order review, so as to reduce illegal medical orders or items that are not reimbursed.

8. A system for collecting, analyzing, and monitoring inpatient medical behavior according to claim 1, characterized in that, The nursing record module includes a signature and review unit, which supports nursing staff to automatically complete electronic signatures and record reviews via voice or commands. The signature and audit unit allows nursing staff to trigger the electronic signature process through voice commands or instructions when performing infusions, changing dressings, special operations, or high-risk medications. The system binds the signature event with the corresponding operation record and timestamp and writes it into the audit log chain to achieve traceability and non-repudiation of nursing behavior.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the control and data processing logic of the system according to any one of claims 1 to 8.

Images