Medical record classification early warning method based on artificial intelligence

Abstract

The invention discloses a medical record file classification early warning method based on artificial intelligence, which belongs to the technical field of artificial intelligence, and comprises the steps of data acquisition, standardization processing, primary and secondary classification prediction, data fusion, early warning judgment and report generation. The technical problems of real-time early warning of potential hospitalization demands in medical records and effective early warning of hospitalization resource shortage are solved, historical data and model prediction are fully utilized, final decisions of doctors are considered, the early warning result is more reasonable and effective, long-term retention prediction information is prevented from affecting early warning, timely data conversion is guaranteed, and the early warning efficiency is improved. According to the method, data acquisition, model calculation, data fusion and early warning report generation are carried out, the whole process can be recorded and traced, and a complete basis is provided for manual intervention, system tuning and subsequent model iteration.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence, and particularly relates to a method for classifying and warning medical record files based on artificial intelligence. Background Art

[0002] At present, hospital systems widely use electronic medical record systems to record patient visits, diagnoses, and treatment information. Traditional electronic medical record management systems mainly rely on manual entry and statistical analysis by doctors to provide support for diagnosis, treatment, and subsequent management.

[0003] Currently, some research and systems attempt to assist in the analysis of medical records through data mining and artificial intelligence algorithms, such as using machine learning to predict disease risks or assist in diagnosis, in order to provide decision-making references for clinical practice.

[0004] However, the allocation of inpatient resources has always been a major problem in hospital management. Existing methods mainly rely on historical data statistics, simple threshold judgments, and experience management, and often respond insufficiently timely or with low accuracy in terms of bed resource allocation and emergency warning.

[0005] In existing systems, outpatient electronic medical records and inpatient registration information are usually scattered in different information systems, lacking effective data fusion and dynamic association, resulting in insufficient real-time performance and accuracy in predicting inpatient demand and bed resource usage status.

[0006] Current warning methods often only rely on statistical analysis and fixed threshold judgments, without fully combining the judgment information entered by doctors, disease characteristics, and time factors for comprehensive prediction, and there are problems of lag and low sensitivity in the warning results.

[0007] Traditional methods mainly rely on manual statistics and analysis of historical data, lacking the ability to dynamically predict real-time data using artificial intelligence models, and unable to effectively capture fluctuations in inpatient demand, resulting in insufficient warning when bed resources are short or premature warning causing resource waste. Summary of the Invention

[0008] The purpose of the present invention is to provide a method for classifying and warning medical record files based on artificial intelligence, which solves the technical problem of real-time warning of potential inpatient demand in medical record files and effectively warning of inpatient resource shortages through multi-source data fusion and phased dynamic prediction mechanisms.

[0009] To achieve the above purpose, the present invention adopts the following technical solutions:

[0010] A method for classifying and warning medical record files based on artificial intelligence, comprising the following steps:

[0011] Step 1: The outpatient electronic case collection module retrieves the electronic cases of each patient from the outpatient information system in real time, extracts the disease characteristics or hospitalization confirmation information, and performs standardization processing to obtain case data;

[0012] The inpatient registration data collection module regularly retrieves the inpatient registration records and bed allocation data from the inpatient management system, and after preprocessing, obtains the inpatient registration data set;

[0013] Step 2: The primary classification processing module retrieves the case data, establishes a logistic regression model, and calculates the hospitalization prediction probability P of each case data based on the disease characteristics 1,i ; Determine whether hospitalization is required according to the preset probability threshold θ1: If yes, attach the time stamp t i , and mark the status flag as "to be converted", generating a predicted hospitalization record; Aggregate the predicted hospitalization records to generate a primary classification record set Γ, including patient ID, prediction probability, time stamp, and status flag;

[0014] Step 3: The secondary classification processing module retrieves the case data and determines whether there is hospitalization confirmation information: If yes, use the hospitalization confirmation information as the determination condition to generate a secondary classification record, including patient ID and hospitalization confirmation time;

[0015] Step 4: The data fusion module regularly scans the primary classification record set Γ and simultaneously obtains the local time t current ; Calculate the validity of each predicted hospitalization record, mark the records that have not been converted after the timeout as invalid, and output the valid primary classification records; Associate and deduplicate the valid primary classification records and secondary classification records by patient ID, and output the fused predicted hospitalization demand number and hospitalization detailed list;

[0016] Step 5: The early warning processing module retrieves the predicted hospitalization demand number and the inpatient registration data set, calculates the bed shortage value, compares it with the preset threshold δ, and generates and outputs the early warning trigger record and statistical data;

[0017] Step 6: The early warning report generation module retrieves the output result of Step 5, generates a detailed early warning report, and displays it in the form of charts, data, or text.

[0018] Preferably, when performing Step 1, it specifically includes the following steps:

[0019] Step 1-1: The outpatient electronic case collection module retrieves the electronic cases of each patient from the outpatient information system in real time. The electronic case contains the disease description and hospitalization confirmation information entered by the doctor, and also includes patient ID, visit date and time, disease description, diagnosis record, and hospitalization confirmation information;

[0020] Step 1-2: Process the text data in the electronic medical records using NLP methods, extract disease characteristics, standardize the formats of each field in the disease characteristics, convert them into a unified coding format, and output the standardized case data records, i.e., case data;

[0021] Step 1-3: Store the standardized case data in a database;

[0022] Step 1-4: The inpatient registration data collection module regularly obtains the latest inpatient registration records and bed allocation data from the inpatient management system. The fields of the data include patient ID, registration time, actual occupied bed information, total number of beds, and current available number of beds;

[0023] Step 1-5: Clean the inpatient registration records and bed allocation data to obtain data with unified and complete field formats, and collect these data to obtain the inpatient registration data set.

[0024] Preferably, when performing Step 2, the specific steps are as follows:

[0025] Step 2-1: Extract feature x from the case data i，j , where i represents the i-th case, x represents the feature value, and j represents the type of feature, specifically the frequency of disease keywords, the value of key indicators, and the diagnostic description indicators;

[0026] Step 2-2: Construct a logistic regression model, use historical case data to train the logistic regression model, and calculate the hospitalization prediction probability for each case. The logistic regression model is as follows:

[0027]

[0028] Among them, P 1,i represents the hospitalization prediction probability of the i-th case data; x i，j represents the value of the j-th feature of the i-th case; w j represents the weight of the j-th feature, and the weight value is a preset value formulated according to the preset hospitalization rules; b represents the bias term; exp(·) represents the natural exponential function;

[0029] Step 2-3: Preset a probability threshold θ1 and judge: If P 1,i ≥ θ1, then the case data to which P 1,i belongs is marked as "requiring hospitalization"; otherwise, it is not marked;

[0030] Step 2-4: For the case data marked as "requiring hospitalization", record the corresponding patient ID and the hospitalization prediction probability P 1,i , and append a timestamp t i , and mark the status as "to be converted" to generate a predicted hospitalization record;

[0031] Step 2-5: Aggregate all predicted hospitalization records to generate a set Γ of first-classification records. The data format of each piece of data in set Γ is: patient ID + predicted probability P 1,i + timestamp t i + status flag.

[0032] Preferably, when executing Step 3, it specifically includes the following steps:

[0033] Step 3-1: The secondary classification processing module retrieves the case data and extracts the hospitalization confirmation information entered by the doctor: if there is hospitalization confirmation information, extract it; if not, mark it as empty;

[0034] The hospitalization confirmation information includes the flag for confirmed hospitalization and the confirmation time;

[0035] Step 3-2: For the case data with hospitalization confirmation information, generate secondary classification records according to the hospitalization confirmation information. The secondary classification records include the patient ID and the hospitalization confirmation time;

[0036] Step 3-3: Aggregate all secondary classification records to obtain and output the set of secondary classification records.

[0037] Preferably, when executing Step 4, the specific steps are as follows:

[0038] Step 4-1: Periodically scan the set Γ of first-classification records to obtain the local time t current ;

[0039] Step 4-2: Calculate each first-classification record to determine whether it is within the preset valid time T. The specific calculation formula is as follows:

[0040]

[0041] where, indicator function, which is 1 if the condition in the formula is satisfied, otherwise 0; t i is the timestamp of the first-classification record;

[0042] Step 4-3: According to the calculation result in Step 2, mark the status flags of all overdue first-classification records as "invalid", and mark the status flags of all first-classification records equal to 1 as "valid", and retain the first-classification records marked as "valid";

[0043] Step 4-4: Associate all the first-classification records marked as "valid" with the set of secondary classification records according to the patient ID, so that the hospitalization prediction requirements of the same patient are only counted once. Specifically, it includes the following steps:

[0044] Step 4-4-1: Define the number of valid primary classification records N1 eff , specifically expressed as:

[0045]

[0046] Step 4-4-2: Obtain the total number N2 of secondary classification records in the secondary classification record set;

[0047] Step 4-4-3: Calculate the total predicted hospitalization demand N after fusion pred , specifically as:

[0048]

[0049] Step 4-4-4: Output the total predicted hospitalization demand N pred , that is, the predicted hospitalization demand number, and at the same time output the hospitalization detailed list composed of patient ID, predicted probability P 1,i and hospitalization confirmation information.

[0050] Preferably, when performing Step 5, it specifically includes the following steps:

[0051] Step 5-1: Obtain the predicted hospitalization demand number and the hospitalization registration data set, obtain the total predicted hospitalization demand N pred and the hospitalization detailed list according to the predicted hospitalization demand number, and obtain the current available bed number B according to the hospitalization registration data set;

[0052] Step 5-2: Calculate the bed shortage value S:

[0053] S = N pred - B;

[0054] where S represents the difference between the predicted hospitalization demand and the current available beds;

[0055] Step 5-3: Set the warning threshold δ, where δ is used to represent the absolute shortage number, and judge the warning condition:

[0056] If S ≥ δ, trigger a warning and generate a warning record; otherwise, do not trigger and do not generate a warning record;

[0057] Step 5-4: Save the warning record, the total predicted hospitalization demand N pred , the current available bed number B, and the bed shortage value S, package and generate a warning trigger record and statistical data, and output.

[0058] A method for classifying and warning medical record files based on artificial intelligence according to the present invention solves the technical problem of real-time warning of potential hospitalization needs in medical record files and effectively warning of shortages of hospitalization resources through multi-source data fusion and a phased dynamic prediction mechanism. The present invention adopts a primary classification and a secondary classification method, which not only makes full use of historical data and model prediction, but also takes into account the final decision of doctors, making the warning result more reasonable and effective. By setting a preset time threshold for the primary classification result, the effectiveness of the prediction data is judged to avoid the influence of long-term retained prediction information on the warning and ensure the timely conversion of data. A mathematical model is established based on the predicted hospitalization needs and real-time bed resource data to clarify the criteria for triggering a warning and ensure that the risks of hospitalization resources are timely feedback. From data collection, model calculation, data fusion to the generation of warning reports, the whole process of the present invention can be recorded and traced, providing a complete basis for manual intervention, system optimization and subsequent model iteration. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 is the main flow chart of the present invention;

[0060] Figure 2 is the flow chart of step 1 of the present invention;

[0061] Figure 3 is the flow chart of steps 2 and 3 of the present invention;

[0062] Figure 4 is the flow chart of step 4 of the present invention;

[0063] Figure 5 is the flow chart of step 5 of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0064] As Figures 1 - 5 shown, a method for classifying and warning medical record files based on artificial intelligence includes the following steps:

[0065] Step 1: The outpatient electronic case collection module obtains the electronic cases of each patient from the outpatient information system in real time, extracts the disease characteristics or hospitalization confirmation information, and performs standardization processing to obtain case data;

[0066] The inpatient registration data collection module regularly obtains the inpatient registration records and bed allocation situation data from the inpatient management system, and after preprocessing, obtains the inpatient registration data set;

[0067] In this embodiment, the basic information usually included in electronic medical records includes the condition description, preliminary diagnosis, and hospitalization confirmation information entered by doctors. In this embodiment, natural language processing (NLP) algorithms are used to segment the text in the electronic medical records, extract keywords, and standardize them to ensure consistent data formats and unified coding, facilitating subsequent model processing. Since natural language processing (NLP) algorithms are prior art, they will not be described in detail.

[0068] When performing step 1, it specifically includes the following steps:

[0069] Step 1-1: The outpatient electronic medical record collection module retrieves the electronic medical records of each patient in real time from the outpatient information system. The electronic medical records contain the condition description and hospitalization confirmation information entered by doctors, and also include the patient ID, visit date and time, condition description, diagnosis record, and hospitalization confirmation information;

[0070] For example: The data fields may include:

[0071] Patient ID: PID12345;

[0072] Visit time: 2025-04-10 09:30;

[0073] Condition description: The patient has symptoms of high fever, cough, and fatigue;

[0074] Diagnosis record: Suspected infection;

[0075] Hospitalization confirmation information (optional): None or confirmed.

[0076] Step 1-2: For the text data in the electronic medical records, use NLP methods for processing, extract disease characteristics, standardize the formats of each field in the disease characteristics, convert them into a unified coding format, and output the standardized case data record, that is, the case data;

[0077] In this embodiment, when using NLP methods for processing, it includes processing such as word segmentation, part-of-speech tagging, and medical thesaurus matching, and finally extracts key information such as disease keywords or diagnosis description indicators.

[0078] Standardization processing can be performed using technologies such as ICD coding and LOINC in the prior art.

[0079] The data format after its processing is as follows:

[0080] After standardization, the condition description "high fever, cough" may be mapped to: {"heat": 39.5, "cough frequency": "high",...}.

[0081] Step 1-3: Store the standardized case data into the database; in this embodiment, a NoSQL database is used for storage to ensure query efficiency and data security.

[0082] Step 1-4: The inpatient registration data collection module regularly obtains the latest inpatient registration records and bed allocation situation data from the inpatient management system. The fields of the data include patient ID, registration time, actual occupied bed information, total number of beds, and current available number of beds.

[0083] For example, the data fields can be:

[0084] Patient ID: PID12345;

[0085] Registration time: 2025-04-10 10:00;

[0086] Actual occupied bed: Bed number B12;

[0087] Total number of beds: 200;

[0088] Current available number of beds: 50.

[0089] Step 1-5: Clean the inpatient registration records and bed allocation situation data to obtain data with unified and complete field formats, and aggregate these data to obtain the inpatient registration data set.

[0090] In this embodiment, the cleaning of the data mainly includes preprocessing such as removing duplicates, filling missing values, and format conversion.

[0091] Step 2: The primary classification processing module retrieves the case data, establishes a logistic regression model, and calculates the hospitalization prediction probability P of each case data according to the disease characteristics. 1,i ; Determine whether hospitalization is required according to the preset probability threshold θ1: if yes, append the timestamp t i , and mark the status flag as "to be converted", generating a predicted hospitalization record; aggregate the predicted hospitalization records to generate the primary classification record set Γ, including patient ID, prediction probability, timestamp, and status flag.

[0092] In this embodiment, the purpose of the primary classification processing module is to perform preliminary analysis and prediction on the standardized case data, calculate the hospitalization prediction probability of each case using the logistic regression model, and perform preliminary classification based on the preset probability threshold. The prediction results are appended with a timestamp and a status flag indicating "to be converted" to prepare for the judgment of the subsequent time window.

[0093] The main situation is: after the doctor gives the order for the patient to be hospitalized, the patient fails to comply with the order to decide to be hospitalized, and at this time, the doctor fails to fill in (or select) the hospitalization information in the electronic case.

[0094] When performing Step 2, the specific steps are as follows:

[0095] Step 2-1: Extract feature x from the case data i，j , where i represents the i-th case, x represents the feature value, and j represents the type of feature, specifically the frequency of disease keywords, the value of key indicators, and the diagnostic description indicators;

[0096] x i，j The examples of

[0097] For case PID12345: Extract the feature vector:

[0098] x i，j ={x i，1 = degree of high fever, x i，2 = cough frequency, x i，3 = abnormal blood routine indicators}.

[0099] x i，j is numerical or discrete encoded data.

[0100] Step 2-2: Construct a logistic regression model, use historical case data to train the logistic regression model, and calculate the hospitalization prediction probability for each case. The logistic regression model is as follows:

[0101]

[0102] where P 1,i represents the hospitalization prediction probability of the i-th case data; x i，j represents the value of the j-th feature of the i-th case; w j represents the weight of the j-th feature, and this weight value is a preset value determined according to the preset hospitalization rules; b represents the bias term; exp(·) represents the natural exponential function;

[0103] Step 2-3: Preset a probability threshold θ1 and judge: If P 1,i ≥θ1, then the case data to which P 1,i belongs is marked as "requiring hospitalization"; otherwise, it is not marked;

[0104] Step 2-4: For the case data marked as "requiring hospitalization", record the corresponding patient ID and the hospitalization prediction probability P 1,i , and append the timestamp t i , and mark the status as "to be converted" to generate a predicted hospitalization record;

[0105] The predicted hospitalization record is as follows:

[0106] {"PID":"PID12345","P":0.68,"t":"2025-04-10 09:32","Status":"To be converted"}。

[0107] Step 2-5: Aggregate all predicted hospitalization records to generate a set Γ of first-classification records. The data format of each piece of data in set Γ is: patient ID + predicted probability P 1,i + timestamp t i + status flag.

[0108] Step 3: The secondary classification processing module retrieves the case data and determines whether there is hospitalization confirmation information. If yes, use the hospitalization confirmation information as the determination condition to generate secondary classification records, including the patient ID and the hospitalization confirmation time;

[0109] When performing Step 3, it specifically includes the following steps:

[0110] Step 3-1: The secondary classification processing module retrieves the case data and extracts the hospitalization confirmation information entered by the doctor. If there is hospitalization confirmation information, extract it; if not, mark it as empty;

[0111] The hospitalization confirmation information includes the confirmation flag for hospitalization and the confirmation time;

[0112] The hospitalization confirmation information is as follows:

[0113] For case PID12345: If the doctor enters a hospitalization decision during the second visit, record it as:

[0114] {"PID":"PID12345","Hospitalization confirmation time":"2025-04-10 10:00"}。

[0115] Step 3-2: For the case data with hospitalization confirmation information, generate secondary classification records according to the hospitalization confirmation information. The secondary classification records include the patient ID and the hospitalization confirmation time;

[0116] Step 3-3: Aggregate all secondary classification records to obtain and output the set of secondary classification records.

[0117] Step 4: The data fusion module periodically scans the set Γ of first-classification records and simultaneously obtains the local time t current ; Calculate the validity of each predicted hospitalization record, mark the records that have not been converted after the timeout as invalid, and output the valid first-classification records; Associate and deduplicate the valid first-classification records and secondary classification records by patient ID for fusion, and output the fused predicted hospitalization demand number and the detailed hospitalization list;

[0118] When performing Step 4, the specific steps are as follows:

[0119] Step 4-1: Regularly scan the classification record set Γ to obtain the local time t current ;

[0120] Step 4-2: Calculate each first-classification record to determine whether it is within the preset valid time T. The specific calculation formula is as follows:

[0121]

[0122] where is an indicator function, which is 1 if the conditions in the formula are met, otherwise it is 0; t i is the timestamp of the first-classification record;

[0123] Step 4-3: According to the calculation results in Step 2, mark the status flags of all overdue first-classification records as "invalid", and mark the status flags of all first-classification records equal to 1 as "valid", and retain the first-classification records marked as "valid"; Mark the status flags of all first-classification records equal to 1 as "valid", and retain the first-classification records marked as "valid";

[0124] Step 4-4: Associate all the first-classification records marked as "valid" with the second-classification record set according to the patient ID, so that the same patient is only counted once for the hospitalization prediction demand, which specifically includes the following steps:

[0125] Step 4-4-1: Define the number N1 of valid first-classification records eff , which is specifically expressed as:

[0126]

[0127] Step 4-4-2: Obtain the total number N2 of second-classification records in the second-classification record set;

[0128] Step 4-4-3: Calculate the total number N of the predicted hospitalization demand after fusion pred , which is specifically:

[0129]

[0130] Step 4-4-4: Output the total number N of the predicted hospitalization demand pred , that is, the predicted hospitalization demand number, and at the same time output the hospitalization detailed list composed of the patient ID, the predicted probability P 1,i and the hospitalization confirmation information.

[0131] Step 5: The early warning processing module retrieves the predicted hospitalization demand number and the hospitalization registration data set, calculates the bed shortage value, compares it with the preset threshold δ, and generates and outputs the early warning trigger record and statistical data;

[0132] When performing Step 5, it specifically includes the following steps:

[0133] Step 5-1: Obtain the predicted number of hospitalization requirements and the hospitalization registration dataset, and obtain the total predicted number of hospitalization requirements N based on the predicted number of hospitalization requirements pred and the detailed hospitalization list. Based on the hospitalization registration dataset, obtain the current available number of beds B;

[0134] Step 5-2: Calculate the bed shortage value S:

[0135] S = N pred - B;

[0136] where S represents the difference between the predicted hospitalization demand and the current available beds;

[0137] Step 5-3: Set the warning threshold δ, where δ is used to represent the absolute shortage number, and judge the warning condition:

[0138] If S ≥ δ, then trigger a warning and generate a warning record; otherwise, do not trigger and do not generate a warning record;

[0139] Step 5-4: Save the warning record, the total predicted number of hospitalization requirements N pred the current available number of beds B, and the bed shortage value S, package them to generate a warning trigger record and statistical data, and output.

[0140] Step 6: The warning report generation module retrieves the output result of Step 5, generates a detailed warning report, and displays it in the form of charts, data, or text.

[0141] In this embodiment, when finally presenting, a detailed report can be generated through a reporting tool (such as a BI system, a chart library), and the content includes numerical statistics, trend charts, and key data summaries.

[0142] In this embodiment, the outpatient electronic case collection module and the inpatient department registration data collection module are both deployed in the data collection layer, and electronic case data is collected from the outpatient system or registration and bed data is collected from the inpatient department management system in real time through the API or HL7 protocol.

[0143] The database is deployed in the data storage layer and is used to store system-level logs, warning records, and feedback data for subsequent analysis and model optimization.

[0144] The primary classification processing module, the secondary classification processing module, the data fusion module, and the warning processing module are all deployed in the analysis and processing layer.

[0145] The warning report generation module is deployed in the report display layer.

[0146] Each layer communicates with each other through the API gateway.

[0147] A method for classifying and warning medical record files based on artificial intelligence solves the technical problem of real-time warning of potential hospitalization needs in medical record files through multi-source data fusion and a phased dynamic prediction mechanism, and effectively warns of the shortage of hospitalization resources. The present invention adopts the methods of primary classification and secondary classification, which not only makes full use of historical data and model prediction, but also takes into account the final decision of doctors, making the warning result more reasonable and effective. By setting a preset time threshold for the primary classification result, the validity of the prediction data is judged to avoid the influence of long-term retained prediction information on the warning and ensure the timely conversion of data. A mathematical model is established based on the predicted hospitalization needs and real-time bed resource data to clarify the criteria for triggering a warning and ensure that the risk of hospitalization resources is timely feedback. From data collection, model calculation, data fusion to the generation of warning reports, the whole process of the present invention can be recorded and traced, providing a complete basis for manual intervention, system optimization and subsequent model iteration.

Claims

1. An artificial intelligence-based medical record classification and early warning method, characterized in that: It includes the following steps: Step 1: The outpatient electronic case collection module retrieves the electronic cases of each patient from the outpatient information system in real time, extracts the disease characteristics or hospitalization confirmation information, and performs standardization processing to obtain case data; The inpatient department registration data collection module regularly retrieves the inpatient registration records and bed allocation data from the inpatient department management system, and after preprocessing, obtains the inpatient registration data set; Step 2: The primary classification processing module retrieves case data, establishes a logistic regression model, and calculates the hospitalization prediction probability P for each case data based on the disease condition characteristics 1,i ; Determine whether hospitalization is required according to the preset probability threshold θ1: if yes, append the timestamp t i , and mark the status flag as "to be converted", generating a predicted hospitalization record; Aggregate the predicted hospitalization records to generate a primary classification record set Γ, including patient ID, prediction probability, timestamp, and status flag; Step 3: The secondary classification processing module retrieves the case data to determine whether there is hospitalization confirmation information. If so, use the hospitalization confirmation information as the determination condition to generate secondary classification records, including patient ID and hospitalization confirmation time; Step 4: The data fusion module scans the classification record set Γ at regular intervals and obtains the local time t simultaneously current ; calculate the validity of each predicted hospitalization record, mark the records that have not been converted after the timeout as invalid, and output the valid first-classification records; associate and deduplicate the valid first-classification records and second-classification records by patient ID, and output the fused predicted hospitalization demand number and the detailed hospitalization list; Step 5: The early warning processing module retrieves the predicted hospitalization demand number and the inpatient registration data set, calculates the bed shortage value, compares it with the preset threshold δ, and generates and outputs the early warning trigger record and statistical data; Step 6: The early warning report generation module retrieves the output result of Step 5, generates a detailed early warning report, and displays it in the form of charts, data or text.

2. The method for classifying and warning medical record files based on artificial intelligence according to claim 1, wherein: When executing Step 1, it specifically includes the following steps: Step 1-1: The outpatient electronic case collection module retrieves the electronic cases of each patient from the outpatient information system in real time. The electronic case contains the disease description and hospitalization confirmation information entered by the doctor, and also includes patient ID, visit date and time, disease description, diagnosis record, hospitalization confirmation information; Step 1-2: Use the NLP method to process the text data in the electronic case, extract the disease characteristics, standardize the format of each field in the disease characteristics, convert it into a unified coding format, and output the standardized case data record, that is, the case data; Step 1-3: Store the standardized case data in the database; Step 1-4: The inpatient department registration data collection module regularly retrieves the latest inpatient registration records and bed allocation data from the inpatient department management system. The fields of the data include patient ID, registration time, actual occupied bed information, total number of beds and current available number of beds; Step 1-5: Clean the inpatient registration records and bed allocation data to obtain data with a unified and complete field format, and aggregate these data to obtain the inpatient registration data set.

3. The method for classifying and warning medical record files based on artificial intelligence according to claim 1, wherein: When executing Step 2, the specific steps are as follows: Step 2-1: Extract feature x from the case data i，j , where i represents the i-th case, x represents the feature value, and j represents the type of feature, specifically the frequency of disease keywords, the value of key indicators, and the diagnostic description indicators; Step 2-2: Construct a logistic regression model, use historical case data to train the logistic regression model, and calculate the hospitalization prediction probability of each case. The logistic regression model is as follows: Among them, P 1,i represents the hospitalization prediction probability of the i-th case data; x i，j represents the value of the j-th feature of the i-th case; w j represents the weight of the j-th feature, and the weight value is a preset value formulated according to the preset hospitalization rules; b represents the bias term; exp(·) represents the natural exponential function; Step 2-3: Preset a probability threshold θ1 and judge: If P 1,i ≥ θ1, then the case data to which P 1,i belongs is marked as "requiring hospitalization"; otherwise, it is not marked. Step 2-4: For the case data marked as "requiring hospitalization", record the corresponding patient ID and the hospitalization prediction probability P 1,i , and append the timestamp t i , mark the status as "to be converted", and generate a predicted hospitalization record; Step 2-5: Aggregate all predicted hospitalization records to generate a first-classification record set Γ, and the data format of each piece of data in set Γ is: patient ID + predicted probability P 1,i + timestamp t i + status flag.

4. The method for classifying and warning medical record files based on artificial intelligence according to claim 1, characterized in that: When executing Step 3, it specifically includes the following steps: Step 3-1: The secondary classification processing module retrieves the case data and extracts the hospitalization confirmation information entered by the doctor. If there is hospitalization confirmation information, extract it; if not, mark it as empty; The hospitalization confirmation information includes the identification of confirmed hospitalization and the confirmation time; Step 3-2: For the case data with hospitalization confirmation information, generate secondary classification records according to the hospitalization confirmation information. The secondary classification records include patient ID and hospitalization confirmation time; Step 3-3: Aggregate all secondary classification records to obtain and output the secondary classification record set.

5. The method for classifying and warning medical record files based on artificial intelligence according to claim 4, characterized in that: When executing Step 4, the specific steps are as follows: Step 4-1: Scan the classification record set Γ at regular intervals to obtain the local time t current ; Step 4-2: Calculate each first-classification record to determine whether it is within the preset valid time T. The specific calculation formula is as follows: I i valid = I((t current - t i ) ≤ T); where I i valid is an indicator function, which is 1 if the condition in the formula is satisfied, and 0 otherwise; t i is the timestamp of a primary classification record; Step 4-3: According to the calculation result of Step 2, mark the status flags of all one-time classification records that have timed out as "invalid", and mark the status flags of all one-time classification records where I i valid is equal to 1 as "valid", and retain the one-time classification records marked as "valid". Step 4-4: Associate all the first-classification records marked as "valid" with the second-classification record set according to the patient ID, so that the hospitalization prediction requirements of the same patient are only counted once. The specific steps are as follows: Step 4-4-1: Define the number of valid first-level classification records N1 eff , specifically expressed as: Step 4-4-2: Obtain the total number N2 of the second-classification records in the second-classification record set; Step 4-4-3: Calculate the total number N of predicted hospitalization demands after fusion pred , specifically as follows: N pred = N1 eff + N2; Step 4-4-4: Output the total predicted hospitalization demand N pred , that is, the predicted number of hospitalization demands, and at the same time output the detailed hospitalization list composed of patient ID, predicted probability P 1,i and hospitalization confirmation information.

6. The method for classifying and warning medical record files based on artificial intelligence according to claim 5, characterized in that: When performing Step 5, the specific steps are as follows: Step 5-1: Obtain the predicted number of hospitalizations and the hospitalization registration dataset, and obtain the total predicted number of hospitalizations N based on the predicted number of hospitalizations; pred and the detailed hospitalization list, and obtain the current available number of beds B based on the hospitalization registration dataset; Step 5-2: Calculate the bed shortage value S: S = N pred -B; Among them, S represents the difference between the predicted hospitalization demand and the current available beds; Step 5-3: Set the warning threshold δ, where δ is used to represent the absolute shortage number, and judge the warning condition: If S≥δ, a warning is triggered and a warning record is generated; otherwise, no warning is triggered and no warning record is generated. Step 5-4: Save the early warning record and the total predicted hospitalization demand N pred , the current available bed number B, and the bed shortage value S, package them to generate an early warning trigger record and statistical data, and output them.

Images