Artificial intelligence-based method for patient screening and dynamic follow-up management in clinical trials

Abstract

The present invention relates to the field of medical information technology, specifically to an artificial intelligence-based method for screening and dynamic follow-up management of clinical trial patients, comprising the following steps: collecting and standardizing admission, examination, medication, and imaging data, mapping trial items to form associated health data, screening qualified patients according to the inclusion criteria, extracting populations with differences in medical unit and patient distribution markers, analyzing physiological data fluctuation trends, adjusting the follow-up sequence to generate a dynamic management plan. In the present invention, by collecting structured health information and mapping trial items, accurate screening and synchronous rhythm control are achieved, the distribution of patients and medical units is analyzed, differential groups are identified, resource allocation is optimized, physiological data is constructed into a time series and the difference is calculated to capture health fluctuation trends. The time schedule is adjusted in conjunction with the follow-up nodes to improve execution efficiency, and multiple links collaborate to enhance the response speed of patient management and the integrity of data processing.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to a method for screening and dynamic follow-up management of clinical trial patients based on artificial intelligence. Background Art

[0002] The field of medical information technology encompasses the collection, storage, processing, analysis, and management of medical data. Core areas within this field include the digitization of patient health data, the interconnection of medical systems, the management of electronic medical records, support systems for disease diagnosis and treatment, data management for clinical trials, and patient health monitoring. With the continuous advancement of information technology, its application in clinical decision-making, disease prediction and diagnosis, the development of personalized treatment plans, and health management is becoming increasingly widespread, driving the digital and intelligent transformation of the healthcare industry. Technologies in this field are constantly being refined to enhance the efficiency and quality of medical services.

[0003] Among them, the AI-based clinical trial patient screening and dynamic follow-up management method refers to the optimization of the screening and dynamic management of patients in clinical trials through artificial intelligence technology. The technical matters targeted by this patent subject include the use of patients' health data, historical case records, genetic information, and other relevant data for comprehensive analysis, automatic identification of patient groups that meet clinical trial conditions, and dynamic health management based on individual health status and follow-up requirements. Specifically, by building a patient screening model and a follow-up management model, combined with artificial intelligence algorithms to conduct real-time monitoring of patient information, data analysis, risk assessment, and follow-up reminders, patients' health status during clinical trials is ensured to be managed and tracked in a timely manner.

[0004] Existing medical information technology primarily relies on static data processing rules, lacking flexibility when processing clinical trial data and struggling to adapt to rapidly changing trial conditions and patient status, resulting in an inability to promptly identify eligible patients. The fragmented nature of data sources and the lack of a unified data format also hinder effective data integration, limiting the scope and accuracy of patient screening. The intermittent nature of health management and the lack of real-time monitoring result in insufficient tracking of patient conditions, making it difficult to promptly detect and address health fluctuations, thus impacting the timeliness of medical interventions. Overlapping schedules and irrational resource allocation in the development and implementation of follow-up plans increase the waste of medical resources and can also lead to interruptions in data collection, compromising the continuity and accuracy of clinical trial data. For example, the lack of an effective data integration mechanism prevents timely integration of medical records across institutions, potentially leading to the omission of important medical information and, in turn, compromising the reliability of clinical trial results. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the existing technology and propose an artificial intelligence-based clinical trial patient screening and dynamic follow-up management method.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an artificial intelligence-based clinical trial patient screening and dynamic follow-up management method, comprising the following steps:

[0007] S1: Deploy a data collection channel in the medical data network to collect admission information, examination records, medication and imaging data, extract key information to generate a unified format data table, map the test items, and obtain the test-related health data;

[0008] S2: Based on the test-related health data, the data is compared with the inclusion criteria, and the records matching the criteria are screened out. The mean time interval between adjacent records is calculated, and the data update rhythm is adjusted to obtain the AI-screened qualified patient group.

[0009] S3: Based on the AI-screened qualified patient group, extract the related medical institutions and patient numbers, calculate and classify the record differences, mark the populations with distribution differences, and obtain a dynamic follow-up management monitoring list;

[0010] S4: Based on the dynamic follow-up management monitoring list, extract the continuously recorded physiological data of the patients, construct a sequence by time order, calculate adjacent differences, identify continuous change segments, establish analysis labels, and obtain the health fluctuation trend of the dynamically monitored population;

[0011] S5: Based on the health fluctuation trends of the dynamically monitored population, find the corresponding time nodes for follow-up arrangements, identify time overlaps and misalignments and adjust the order, organize them into update items, and generate an AI-driven dynamic patient management plan.

[0012] The associated health data specifically includes admission information, examination records, medication data, and imaging data. The group of patients qualified in the initial screening by artificial intelligence includes screening condition matching records, time interval data, and data update frequency. The dynamic follow-up management monitoring list includes medical unit identification, number of patients, and distribution difference marks. The health fluctuation trend of the dynamically monitored population specifically refers to time series, health difference, and fluctuation labels. The artificial intelligence-driven patient dynamic management plan includes follow-up time nodes, sequence adjustment details, and update items.

[0013] As a further solution of the present invention, the specific steps of S1 are:

[0014] S101: Deploy a data collection channel in the medical data network to collect medical records, laboratory tests, imaging and admission records, examination items, imaging descriptions, and drug information. Unify the field format, perform field mapping, and normalize data types to obtain a unified field table.

[0015] S102: Performing structural analysis and content recognition based on the field items in the unified field table, screening field records, calculating field completeness, and eliminating records that do not meet the requirements by setting a threshold to obtain standard records of structural fields;

[0016] S103: Based on the standard record of the structure field, the record fields are matched with the experimental design requirements, the fields that meet the requirements are compared and a mapping relationship is established, and combined with the dynamic follow-up management node evaluation, the records that meet the experimental requirements are screened to obtain the experimental associated health data.

[0017] As a further solution of the present invention, the specific steps of S2 are:

[0018] S201: Based on the test-related health data, each field is compared item by item according to the preset inclusion criteria, and field features consistent with the standard content are recorded to generate field matching statistics;

[0019] S202: Based on the field matching statistical results, sort the records that meet the inclusion criteria by reception time, calculate the reception time interval between adjacent records, obtain the time difference between every two records, and generate the average record time interval;

[0020] S203: Based on the mean of the recording time intervals, the synchronization rhythm of the data batch updates is adjusted, the data update frequency is calculated, and the batches are allocated according to the update frequency. The patient records that meet the inclusion criteria are included in the update queue to generate an artificial intelligence initial screening qualified patient group.

[0021] As a further solution of the present invention, the calculation formula of the data update frequency is specifically:

[0022] ;

[0023] in, Represents the data update frequency, Represents the receiving time of the i-th record, Represents the difference between the i-th record and the previous record, Represents the total number of records, Represents the time difference between adjacent records, Represents the absolute value of the difference from the previous record.

[0024] As a further solution of the present invention, the specific steps of S3 are:

[0025] S301: Based on the AI-preliminary screening of qualified patients, extract the name of the medical institution and the corresponding number of patients associated with each record, group the records by medical institution, count the number of patients in each institution, and calculate the difference in the number of patients between institutions to generate a statistical table of the difference in the number of patients per institution;

[0026] S302: Based on the unit patient quantity difference statistics table, partition the quantity difference ranges between units, classify units with deviations in the difference as high-priority monitoring areas, mark the units as areas with high data update frequency, extract the corresponding patient records, and generate a unit difference classification marking table;

[0027] S303: Based on the unit difference classification mark table, the patient records in the high priority area are included in the dynamic follow-up management focus range, a dynamic follow-up management task schedule is generated, and the monitoring list is updated synchronously to obtain a dynamic follow-up management monitoring list.

[0028] As a further solution of the present invention, the calculation formula of the quantity difference between the units is specifically:

[0029] ;

[0030] in, Represents the quantity difference between units, Representative The number of patients in each medical unit, Representative The number of patients in each medical unit, Representative Unit The adjustment value of Representative Unit The adjustment value of .

[0031] As a further solution of the present invention, the specific steps of S4 are:

[0032] S401: Based on the dynamic follow-up management monitoring list, extract the body temperature, blood sugar and blood pressure data of each patient, sort the data in chronological order, and generate a sorted data sequence;

[0033] S402: Based on the sorted data sequence, calculate the numerical difference between adjacent records to obtain the daily change amount, and mark the change direction of the daily data according to the change value, identify records that maintain a consistent trend for multiple days, and generate change trend direction data;

[0034] S403: Based on the change trend direction data, the records showing a consistent change trend over multiple days are screened, and the records with obvious trend characteristics are marked as abnormal fluctuations and summarized to obtain the health fluctuation trend of the dynamically monitored population.

[0035] As a further solution of the present invention, the calculation formula for the numerical difference between adjacent records is specifically:

[0036] ;

[0037] in, Represents the numerical difference between adjacent records. Representative Daily data record value, Represents the data record value of the previous day, Representative The daily weight value, Representative The data weight value of the day.

[0038] As a further solution of the present invention, the specific steps of S5 are:

[0039] S501: Based on the health fluctuation trend of the dynamically monitored population, extract the patient number for which the follow-up plan needs to be adjusted, call the individual patient's scheduling record, search and match the associated follow-up nodes, obtain the time interval of each node, and generate the patient follow-up node time interval;

[0040] S502: Based on the patient follow-up node time interval, compare the node schedule with the preset time, identify the node content with time overlap or inconsistency, record and mark the nodes that need to be adjusted, and generate a time node adjustment mark table;

[0041] S503: Based on the time node adjustment mark table, adjust the trigger order of the nodes with overlap and inconsistency, set the newly added task node identifier, synchronize all updated information to the patient's individual plan table, and generate an artificial intelligence-driven patient dynamic management plan.

[0042] Compared with the prior art, the advantages and positive effects of the present invention are:

[0043] This method collects structured health information and maps test items to achieve precise screening and synchronized rhythm control. It analyzes the distribution of patients and medical institutions, identifies differential groups, and optimizes resource allocation. It constructs time series from physiological data and calculates differences to capture health fluctuation trends. It also adjusts schedules based on follow-up nodes to improve execution efficiency. This multi-step collaboration enhances patient management response speed and data processing integrity. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 Schematic diagram of the steps of the present invention; DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0046] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0047] See also Figure 1 , an artificial intelligence-based clinical trial patient screening and dynamic follow-up management method includes the following steps:

[0048] S1: Establish a data collection channel connecting medical records, tests, imaging, and medication in the medical data network. Collect patient admission information, examination records, medication lists, and imaging descriptions. Extract patient identification, recording time, test values, drug names, and imaging keywords. Write the data into the task registration form in a standardized format. Establish a correspondence between clinical trials and data fields. Identify screening conditions and evaluate dynamic follow-up management nodes to obtain trial-related health data.

[0049] S2: Based on the trial-associated health data, each field is marked for matching with the inclusion criteria one by one. Records with a fixed threshold of matching fields are included in the pass list. The data reception time of the pass records is arranged in sequence, the average time interval between adjacent records is calculated, and the data batch update rhythm is adjusted to obtain the AI-screened qualified patient group.

[0050] S3: Based on the AI-based initial screening of qualified patients, the name of the medical institution associated with each record and the number of corresponding patients are extracted. The difference in the number of units is calculated, and classification and labeling are performed based on the difference range. After classification, the patients corresponding to the units with the most frequent record updates are included in the dynamic follow-up management scheduling focus range, and a dynamic follow-up management monitoring list is generated;

[0051] S4: Based on the dynamic follow-up management monitoring list, extract the recent temperature, blood sugar, and blood pressure records of each patient on the list, construct a continuous data sequence by time, calculate the difference between adjacent values, and mark the daily change direction. Identify data segments that show a consistent change trend over multiple consecutive days, and include records with trend characteristics in the identification results to obtain the health fluctuation trend of the dynamically monitored population;

[0052] S5: Based on the dynamic monitoring of the health fluctuation trends of the population, extract the patient numbers that need to adjust the follow-up plan, call the scheduling records to find the related follow-up nodes, compare the trend data time with the current node time interval, identify the node content with time overlap or inconsistent time sequence, adjust the trigger sequence and set the new node identifier, synchronize all updated information to the patient's individual plan table, and generate an AI-driven patient dynamic management plan.

[0053] The associated health data specifically include admission information, examination records, medication data, and imaging data. The group of patients who pass the initial screening by artificial intelligence includes screening condition matching records, time interval data, and data update frequency. The dynamic follow-up management monitoring list includes medical unit identification, number of patients, and distribution difference marks. The dynamic monitoring of the health fluctuation trends of the population specifically refers to time series, health difference values, and fluctuation labels. The artificial intelligence-driven dynamic patient management plan includes follow-up time nodes, sequence adjustment details, and update items.

[0054] The specific steps of S1 are:

[0055] S101: Deploy a data collection channel in the medical data network to collect medical records, laboratory tests, imaging and admission records, examination items, imaging descriptions, and drug information. Unify the field format, perform field mapping, and normalize data types to obtain a unified field table.

[0056] First, a data collection channel is deployed within the medical data network. Interfaces are used to connect to the medical information system to obtain field data such as patient admission records, laboratory data, imaging results, and medication information. After data collection, it is formatted by establishing a standardized data field structure, ensuring that information provided by different data sources meets a unified formatting standard. By comparing data structures across different medical systems, a unified data table is generated to ensure standardized management and effective integration of subsequent data. For example, patient admission information includes fields such as patient name, admission date, and ward number, while laboratory test items include fields such as blood routine and urine routine. Through data connection and mapping, these fields are unified into a standardized format to ensure that the data can be processed and stored by the system. Furthermore, image descriptions provided by the imaging system may contain complex text information and require normalization using pre-set parsing rules. For example, "chest X-ray shows lung shadows" can be standardized as "chest image shadows." Medication information, such as drug name, dosage, and instructions for use, is standardized through unified field mapping to ensure compatibility. Next, the collected fields are normalized to different data types. For example, date fields are standardized to the YYYY-MM-DD format, and numeric fields are standardized to floating number format. Finally, all normalized data are organized into a unified field table and used as the basis for subsequent screening and analysis. Through this process, a unified field table is obtained.

[0057] S102: Performing structural analysis and content recognition based on the field items in the unified field table, screening field records, calculating field completeness, and eliminating records that do not meet the requirements by setting a threshold to obtain standard records of structural fields;

[0058] First, based on a unified field table, each patient's ID, test data, recording time, medication information, and imaging description are extracted. For example, Patient A's ID is "12345," and their test data includes a blood glucose level of "5.2 mmol / L," medication information of "Amoxicillin," an imaging description of "right lung shadow," and a recording time of "2025-04-08 09:00:00." These fields provide direct access to the patient's basic information and various test and medical data. Next, these fields are filtered to check for data integrity and validity. For example, if the test data is empty or the date field format is not compliant, the record is considered invalid. Next, the field integrity of each record is calculated to check for missing or erroneous data. Field integrity is calculated based on predefined rules (e.g., "all five fields are not empty"). For Patient A, field integrity calculation is used as an example: if both the "Test Data" and "Recording Time" fields meet the requirements, the record is considered 100% complete. If a field is missing (e.g., the test data is empty), the record's completeness is reduced and it ultimately fails the screening. Based on this, a field completeness threshold is set, such as specifying that records with field completeness below 80% will be eliminated. Such records will be marked and removed. Ultimately, all records that pass the screening will be organized and recorded in a new table, forming standard records with structured fields.

[0059] S103: Based on the standard record of structure fields, the record fields are matched with the experimental design requirements, the fields that meet the requirements are compared and a mapping relationship is established, and the records that meet the experimental requirements are screened in combination with the dynamic follow-up management node evaluation to obtain the experimental associated health data;

[0060] First, extract the values ​​of each field in each record. For example, the record time field is extracted as "2025-04-08 10:30:00", the test field value is extracted as "blood sugar = 6.1mmol / L", the image field value is "lung shadow", and the drug field is "amoxicillin". Then, check each field value with the corresponding indicators in the test design requirements one by one. The blood sugar is required to be between 4.4-6.4mmol / L and not empty. The record time must be after the test start date "2025-04-01". The drug field must not contain banned drug keywords such as "aspirin". The image field should identify the preset keywords. For words such as "shadow" or "infiltration", the field comparison operation is completed using the conditional judgment method, that is, the upper and lower limits of the test field range are set to [4.4, 6.4]. If the field value falls within the interval, it is judged to be satisfied. The record time field uses the timestamp comparison method, and the difference operation is performed after converting it to the timestamp with the start date to determine whether it is a positive value. The image keyword uses the matching keyword set and the field description to determine the string inclusion relationship. If there is a keyword match, the condition is met. The key drug exclusion list is called for the drug field, and the matching drug name set and the field value are screened for intersection to determine whether it contains banned drug items. If so, it is judged to be If it is not satisfied, after completing the comparison of each field, the field items that meet the requirements are accumulated through scoring, and the field matching threshold is set to 80% of the total number of fields. For example, when the total number of fields is 10, at least 8 field matching requirements must be met. If a record meets more than 8 field items, it will enter the next step of processing, establish a mapping relationship between field values ​​and test standards, and uniformly map local field names to field names used in test standards through the field label name conversion table, such as "blood sugar value" is unified as "GLU", and then fill in the mapped data structure in the order of fields. After the mapping is established, node evaluation processing is performed to extract the record time field in each record and judge its Which follow-up node evaluation interval does it fall into? The interval is set as 7-day, 14-day, and 30-day nodes according to the test plan. The node evaluation rule is set as if the difference between the recording time and the time of the most recent follow-up node is less than 3 days, then the number of records of this node will be counted. If the difference is negative or greater than 3 days, it will be counted as a deviation record. The threshold is set as the record items that need node evaluation attention when the proportion of records deviating from the node is higher than 30%. For example, if 2 out of 5 patient records deviate from the node, the deviation rate reaches 40%, which is higher than the set threshold. Finally, the records that meet the field comparison requirements and pass the node evaluation range will be screened out as records that meet the test requirements, and the test-related health data will be obtained.

[0061] The specific steps of S2 are:

[0062] S201: Based on the trial-related health data, each field is compared item by item according to the preset inclusion criteria, and the field features consistent with the standard content are recorded to generate field matching statistics;

[0063] First, based on the trial-linked health data, key information is extracted from each record, including the patient's age, gender, diagnosis, and prohibited medications. For example, Patient A's record contains "Age: 60," "Sex: Male," "Diagnosis: Hypertension," and "Prohibited Medications: Aspirin." These fields are then compared against the pre-set inclusion criteria. Assuming the inclusion criteria require age between 40 and 70, gender, diagnosis of hypertension or diabetes, and prohibited medications of aspirin or warfarin, Patient A's record is considered to meet the criteria. This is because the age, gender, and diagnosis all meet the criteria, and the prohibited medication fields are also valid. Next, the number of fields that meet each requirement is recorded. For example, if Patient A's record meets the requirements for three fields (age, gender, and diagnosis), the system will record the number of fields that meet the criteria after the field comparison is complete, which will serve as a basis for further screening. This process ensures that each record is strictly compared against the inclusion criteria and provides data support for subsequent screening and processing.

[0064] The system then screens all records that meet the criteria. Records with all fields complete and meeting the requirements are included in the subsequent analysis, while records that do not meet the criteria in any field are eliminated. For example, if Patient B's record lists an age of 80, which is outside the age range for inclusion, the record will be eliminated. Ultimately, all records that meet the criteria are organized into a single data set, generating field matching statistics.

[0065] S202: Based on the field matching statistics, sort the records that meet the inclusion criteria by reception time, calculate the reception time interval between adjacent records, obtain the time difference between every two records, and generate the average record time interval;

[0066] First, based on the field matching statistics, we sort all records that meet the inclusion criteria by receipt time. For example, if there are three records with the receipt times of patients A, B, and C, respectively, "2025-04-01 10:00:00," "2025-04-01 14:30:00," and "2025-04-02 09:00:00," the system will sort them from earliest to latest based on these receipt times. This step ensures that all data is logically arranged in time, facilitating subsequent operations. Next, the system calculates the receipt time interval between adjacent records. In the example above, the difference in receipt time between records A and B is 4 hours and 30 minutes, and the difference in receipt time between records B and C is 18 hours and 30 minutes. In this step, the time difference between each pair of records is calculated. For each pair of adjacent records, the system subtracts the two receipt times to obtain the corresponding time difference. Next, the system calculates the mean of the time intervals based on these time differences. For example, if the time difference between records A and B is 4 hours and 30 minutes, and the time difference between records B and C is 18 hours and 30 minutes, the average of the two time differences is (4.5 + 18.5) / 2 = 11 hours. This average reflects the average reception time interval between all records and is used for subsequent scheduling optimization. Through these steps, the average recording time interval is finally generated.

[0067] S203: Based on the mean of the recording time intervals, the synchronization rhythm of the data batch updates is adjusted, the data update frequency is calculated, and the batches are allocated according to the update frequency. The patient records that meet the inclusion criteria are included in the update queue to generate the AI-screened qualified patient group;

[0068] The calculation formula for data update frequency is:

[0069] ;

[0070] in, Represents the data update frequency, Represents the receiving time of the i-th record, Represents the difference between the i-th record and the previous record, Represents the total number of records, Represents the time difference between adjacent records, Represents the absolute value of the difference from the previous record;

[0071] Calculation process:

[0072] Assume there are 5 records with time intervals of 15 minutes, 20 minutes, 10 minutes, and 15 minutes, and the differences are 2mg, 4mg, 1mg, and 3mg, respectively. Calculate according to the formula :

[0073] Calculate the time interval:

[0074] minute, minute, minute, minute;

[0075] Calculate the difference:

[0076] mg, mg, mg, mg;

[0077] Calculation adjustment:

[0078] , , , ;

[0079] Calculate update frequency:

[0080] ;

[0081] ;

[0082] ;

[0083] This result indicates the frequency of data update. The value of 14.1 indicates that the update frequency per unit time for the entire record set is 14.1. This value helps optimize data processing strategies. Specifically, this update frequency value can be used to set the update cycle of data batches or adjust the timing of data synchronization, ensuring that the system can balance the update load and effectively synchronize different records. As patient data changes, the update frequency value will also change, thereby adjusting the synchronization strategy to ensure efficient and flexible data processing.

[0084] The specific steps of S3 are:

[0085] S301: Based on the AI-based initial screening of qualified patient groups, extract the name of the medical institution and the corresponding number of patients associated with each record, group the records by medical institution, count the number of patients in each institution, and calculate the difference in the number of patients between institutions to generate a statistical table of the difference in the number of patients per institution;

[0086] The formula for calculating the quantity difference between units is:

[0087] ;

[0088] in, Represents the quantity difference between units, Representative The number of patients in each medical unit, Representative The number of patients in each medical unit, Representative Unit The adjustment value of Representative Unit The adjustment value of

[0089] Calculation steps:

[0090] Hypothetical Unit and units The number of patients are:

[0091] ;

[0092] ;

[0093] At the same time, assuming the unit adjustment values ​​are:

[0094] ;

[0095] ;

[0096] Calculate the difference in the number of patients:

[0097] ;

[0098] ;

[0099] ;

[0100] ;

[0101] ;

[0102] Calculation unit and units The standardized difference between:

[0103] ;

[0104] This result shows that Indicates the unit and units The difference in the number of patients between the two units was 0.189 after standardization. The number of patients between the two units was relatively small. A value close to 0 indicates that the difference in the number of patients between the two units is small, while a value close to 1 indicates that the difference is significant. and It helps to deal with the scale differences in the number of patients in differentiated units, making the results more reasonable and comparable.

[0105] S302: Based on the unit patient quantity difference statistics table, the quantity difference range between units is partitioned, and units with deviations in the difference are divided into high-priority monitoring areas. The units are marked as areas with high data update frequency, and the corresponding patient records are extracted to generate a unit difference classification marking table;

[0106] First, based on the unit patient population difference statistics table, we need to distinguish the patient population differences within each unit. For example, assume there are three medical units, A, B, and C, with patient populations of 100, 150, and 200, respectively. The patient population differences between adjacent units A and B, and between B and C, are 50 and 50, respectively, and the patient population difference between units A and C is 100. The system zoning is performed based on these difference ranges, for example, setting difference intervals of 0-50, 51-100, 101-150, and so on, to assign different units to different intervals based on their patient population differences. Next, the system further categorizes unit population differences using set difference thresholds. In the above example, if the difference is greater than or equal to 50, the unit is classified as a "high-priority monitoring area." If the difference between units A and B, and between units B and C, is less than 50, they are classified as regular monitoring areas. However, if the difference between units A and C exceeds 100, they are classified as high-priority monitoring areas. In this process, the specific setting of the difference threshold is closely related to the difference in the number of patients in the unit and the actual situation, so the threshold setting can be adjusted according to the distribution of actual data. Then, based on the division results of these difference ranges, the high-priority monitoring areas of the unit will be marked as areas with intensive data update frequency, which means that these units require more frequent monitoring and updates. For example, the number of patients in units A and C is quite different, so they will be marked as units that require frequent data updates. This step is of great significance for the subsequent real-time update and adjustment tasks of data, and can ensure the timeliness and accuracy of data updates. Finally, the system extracts the corresponding patient records according to the above steps, further processes and generates a unit difference classification marking table.

[0107] S303: Based on the unit difference classification mark table, patient records in high-priority areas are included in the dynamic follow-up management focus range, a dynamic follow-up management task schedule is generated, and a monitoring list is simultaneously updated to obtain a dynamic follow-up management monitoring list;

[0108] First, patient records classified as high-priority monitoring areas are extracted from the unit difference classification flag table. These patient records originate from units with significant differences (e.g., a significant difference in the number of patients between Unit A and Unit B). Because these records have higher monitoring needs, the system automatically flags them and assigns them to dynamic follow-up management focus. For example, if the difference between Unit A and Unit B exceeds a threshold, the system automatically selects patient data associated with these units and includes them in the next follow-up management task schedule. Next, the system generates a dynamic follow-up management task schedule based on criteria such as the recording time and patient diagnosis for the selected high-priority patient records. The system schedules follow-up tasks for each patient according to pre-set rules, such as the severity of the patient's condition and the duration of the follow-up. For example, Patient A, due to his high-risk assessment, will be scheduled for multiple follow-up visits in a short period of time, while Patient B's tasks will be scheduled over longer intervals. This process ensures that patients of different priority levels receive appropriate follow-up. The system then integrates this generated task schedule with existing patient data to synchronously update the monitoring list. This process updates all patient follow-up schedules and tasks, ensuring that high-priority patients receive timely follow-up. For example, if Patient A's follow-up schedule changes, the system automatically updates the patient's monitoring list and notifies relevant medical staff. Ultimately, all patient data and follow-up tasks will be integrated into the system, generating a dynamic follow-up management monitoring list.

[0109] The specific steps of S4 are:

[0110] S401: Based on the dynamic follow-up management monitoring list, extract the body temperature, blood sugar and blood pressure data of each patient, sort the data in chronological order, and generate a sorted data sequence;

[0111] First, each patient's basic health data, including temperature, blood sugar, and blood pressure, is obtained from the dynamic follow-up management monitoring list. This data is used to continuously track the patient's health status. Specifically, the system extracts the most recent health indicator data, such as temperature, blood sugar, and blood pressure, from each patient's record. For example, suppose Patient A's record shows a temperature of 37.2°C, a blood sugar of 6.1 mmol / L, and a blood pressure of 120 / 80 mmHg. These data are extracted as basic health information. Next, the system sorts this data chronologically. This sorting ensures a continuous time series so that subsequent analysis accurately reflects the patient's health trends. For example, for Patient A, the system sorts the data based on the timestamp (e.g., "April 8, 2025, 08:00") to ensure that each record corresponds to its corresponding date and time, avoiding erroneous analysis caused by data time confusion. After sorting, all data forms a sorted data series. This series contains the changes in temperature, blood sugar, and blood pressure for each patient at different time points. Suppose that patient A's records at different time points are as follows: 08:00 on April 8, 2025: temperature 37.2°C, blood glucose 6.1mmol / L, blood pressure 120 / 80mmHg; and 12:00 on April 8, 2025: temperature 37.4°C, blood glucose 6.4mmol / L, blood pressure 122 / 82mmHg. The system will arrange this data in chronological order and assign an appropriate sequence number to each time point to facilitate subsequent trend analysis or early warning. Through this process, a continuous health data sequence is established for each patient.

[0112] S402: Based on the sorted data sequence, calculate the numerical difference between adjacent records to obtain the daily change amount, and mark the change direction of the daily data according to the change value, identify records that maintain a consistent trend over multiple days, and generate change trend direction data;

[0113] The formula for calculating the numerical difference between adjacent records is:

[0114] ;

[0115] in, Represents the numerical difference between adjacent records. Representative Daily data record value, Represents the data record value of the previous day, Representative The daily weight value, Representative The data weight value of the day.

[0116] Assume that today's body temperature (Unit: °F), yesterday's body temperature The data is automatically collected by the temperature detection equipment at hourly granularity and then averaged daily, with weights set. 、 ;

[0117] Calculate the absolute value of the change:

[0118] ;

[0119] Compute the square root of the product of consecutive values:

[0120] ;

[0121] Calculate the weighted mean part:

[0122] ;

[0123] Substituting the above three parts into the formula, we get:

[0124] ;

[0125] The results show that there is a standardized variation of 0.9 between the current record and the previous record. This value represents the intensity of the impact of temperature changes on the stability of individual vital signs in dynamic trend identification, and can be used to judge whether there is a continuous upward trend and whether the warning threshold has been reached.

[0126] S403: Based on the change trend direction data, records showing consistent change trends over multiple days are screened, and records with obvious trend characteristics are marked as abnormal fluctuations and summarized to obtain the health fluctuation trend of the dynamically monitored population;

[0127] First, the system analyzes the direction of change in indicators like temperature, blood sugar, and blood pressure across multiple days based on trend direction data, and selects records that exhibit consistent trends over multiple days. Specifically, the system examines the fluctuation trends of each patient's temperature, blood sugar, and blood pressure data over several consecutive days. For example, if a patient's temperature shows an upward trend for three consecutive days, or if their blood sugar continues to rise for three consecutive days, these records are considered to have a consistent trend. The key to this process is identifying the direction of data change and comparing data points in chronological order to ensure that only data that show consistent changes over multiple days and in a consistent direction are selected. For example, if Patient A's temperature data is 37.1°C on April 8, 2025, 37.3°C on April 9, 2025, and 37.5°C on April 10, 2025, this continuous record indicates an upward trend in temperature, so these three days of data are identified as records with a consistent trend. Next, the system flags any data that demonstrates a consistent trend as abnormal fluctuations, drawing the doctor's attention. For example, if patient A's blood sugar rises to near-hyperglycemic levels for three consecutive days (e.g., from 5.8mmol / L to 7.2mmol / L), the data point will be marked as abnormal fluctuation data. The system will determine whether it is an abnormal trend by comparing historical data or setting thresholds when marking the data. Finally, the system will summarize all records of obvious trend features to generate a dynamic monitoring of the health fluctuation trends of the population.

[0128] The specific steps of S5 are:

[0129] S501: Based on the health fluctuation trend of the dynamically monitored population, extract the patient number for which the follow-up plan needs to be adjusted, call the individual patient's scheduling record, search and match the associated follow-up nodes, obtain the time interval of each node, and generate the patient follow-up node time interval;

[0130] First, dynamic health data is continuously monitored, and the monitoring time range is divided into multiple 24-hour cycles. The maximum, minimum, average and standard deviation of the fluctuation curves of the patient's vital signs parameters such as heart rate, blood pressure, blood oxygen, body temperature, etc. are recorded in each cycle. By setting the interval judgment standard, for example, if the daily fluctuation amplitude of heart rate exceeds 30 times / minute, the difference between the maximum and minimum values ​​of blood pressure is greater than 40 mmHg, the daily average value of blood oxygen is lower than 92%, and the body temperature exceeds 37.5℃ for three consecutive days, it is marked as abnormal fluctuation. According to the above judgment standard, all individual numbers with abnormal values ​​in any indicator during the monitoring period are extracted as the key objects for which the follow-up rhythm needs to be adjusted. The number list is output as the basic data for subsequent processing. For each individual number in the number list, its corresponding follow-up schedule record file is called to extract the follow-up nodes corresponding to each time point. The extraction process includes indexing the data associated with each number, reading the node field, filtering the node items with the follow-up mark field as "activated", and eliminating the node items marked as "invalid" or "cancel". The planned time field of each valid node is used to calculate the time interval of the node in combination with the historical execution log of the node. If the actual completion time of the node is more than 3 days earlier than the planned time, the time interval is counted as "ahead of schedule". If the actual completion time is no more than 7 days later than the planned time, it is counted as "normal". Otherwise, it is marked as "delayed". In this way, the node execution record status is standardized. During the matching process, all follow-up node records are matched for each number, and a time interval classification label is constructed in the record. Finally, the follow-up node time interval information of each patient is generated and the result table is output in a structured manner. The fields include patient number, node number, planned time, actual time, time interval classification, etc.

[0131] S502: Based on the patient follow-up node time interval, compare the node schedule with the preset time, identify the node content with time overlap or inconsistency, record and mark the nodes that need to be adjusted, and generate a time node adjustment mark table;

[0132] First, the system obtains time information for each patient's follow-up visit, recording the start and end times of each visit. For example, suppose Patient A's follow-up visit period is from April 1, 2025, to April 5, 2025, and the system collects temperature data with timestamps from April 2, 2025, to April 4, 2025. The system compares this data with the follow-up visit time range. The system first compares the trend data timestamps to the follow-up visit time range, ensuring that the data and visit time ranges are aligned to avoid omissions or conflicts. Next, the system uses this data to determine whether there are any overlaps or discrepancies. For example, if a data point has a timestamp outside the follow-up visit range, such as temperature data with a time stamp of April 6, 2025, the system will flag this data point as inconsistent and require further adjustment or rescheduling of the follow-up visit. If multiple data points have time misalignments, the system will collect and flag these discrepancies and record the time ranges that require adjustment. The system marks any nodes that require adjustment and updates the corresponding patient follow-up plan to ensure they align with the new time interval. For example, for Patient B, the system discovers time mismatches between multiple follow-up nodes. The system reschedules and generates a new follow-up plan based on the existing time data. These adjusted nodes are added to an adjustment mark table for further follow-up and analysis. Ultimately, a time node adjustment mark table is generated.

[0133] S503: Adjust the marking table based on the time nodes, adjust the triggering order of the nodes that overlap or do not match, set the newly added task node identifiers, synchronize all updated information to the patient's individual plan table, and generate an AI-driven dynamic patient management plan;

[0134] First, the system adjusts the marking table based on time nodes, retrieves all overlapping or discrepant follow-up nodes, and compares them with the nodes in the original plan. For example, suppose Patient A's follow-up plan contains two overlapping nodes: one originally scheduled for May 10, 2025, to May 12, 2025, and the other for May 10, 2025, to May 13, 2025. The system will identify the overlap and determine a discrepancy between the two nodes. The system will automatically adjust the triggering order of the nodes, rescheduling tasks to avoid time conflicts. During this process, the system compares the time intervals of each node to identify any discrepancies and determines whether the triggering order needs to be adjusted by determining whether the time intervals overlap. For example, if a node occurs within a misaligned time period, the system will reorder and mark the follow-up nodes based on the latest health data and the patient's medical history. If there are still issues with the reordered node triggering order, the system will continue to adjust until all time nodes comply with the rules, ensuring the smooth execution of all follow-up tasks. Next, the system will set a new task node identifier to clearly identify the adjusted node. For example, if the follow-up node of patient B cannot be adjusted and the task time needs to be postponed, the system will generate a new task identifier for the node and mark it as a new node. The system will ensure that the newly added task node does not conflict with other nodes and keeps the patient's health data consistent with the follow-up plan. Finally, the system will synchronize all adjusted task information to the patient's individual plan to ensure that each patient's follow-up task is updated in a timely manner and reflected in the individual's health management plan. All update and adjustment information will be archived in the form of system-generated update records for subsequent reference. Finally, the system will generate an AI-driven patient dynamic management plan based on the adjusted task schedule.

[0135] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. An artificial intelligence-based method for clinical trial patient screening and dynamic follow-up management, characterized in that: The following steps are involved: S1: Deploy a data collection channel in the medical data network to collect admission information, examination records, medication and imaging data, extract key information to generate a unified format data table, map the test items, and obtain the test-related health data; S2: Based on the test-related health data, the data is compared with the inclusion criteria, and the records matching the criteria are screened out. The mean time interval between adjacent records is calculated, and the data update rhythm is adjusted to obtain the AI-screened qualified patient group. S3: Based on the AI-screened qualified patient group, extract the related medical institutions and patient numbers, calculate and classify the record differences, mark the populations with distribution differences, and obtain a dynamic follow-up management monitoring list; S4: Based on the dynamic follow-up management monitoring list, extract the continuously recorded physiological data of the patients, construct a sequence by time order, calculate adjacent differences, identify continuous change segments, establish analysis labels, and obtain the health fluctuation trend of the dynamically monitored population; S5: Based on the health fluctuation trends of the dynamically monitored population, find the corresponding time nodes for follow-up arrangements, identify time overlaps and misalignments, adjust the order, organize them into update items, and generate an AI-driven dynamic patient management plan; The specific steps for dynamically monitoring the health fluctuation trends of the population are: S401: Based on the dynamic follow-up management monitoring list, extract the body temperature, blood sugar and blood pressure data of each patient, sort the data in chronological order, and generate a sorted data sequence; S402: Based on the sorted data sequence, calculate the numerical difference between adjacent records to obtain the daily change amount, and mark the change direction of the daily data according to the change value, identify records that maintain a consistent trend for multiple days, and generate change trend direction data; S403: Based on the change trend direction data, records showing a consistent change trend over multiple days are screened, and records with obvious trend characteristics are marked as abnormal fluctuations and summarized to obtain the health fluctuation trend of the dynamically monitored population; The specific steps of S1 are: S101: Deploy a data collection channel in the medical data network to collect medical records, laboratory tests, imaging and admission records, examination items, imaging descriptions, and drug information. Unify the field format, perform field mapping, and normalize data types to obtain a unified field table. S102: Performing structural analysis and content recognition based on the field items in the unified field table, screening field records, calculating field completeness, and eliminating records that do not meet the requirements by setting a threshold to obtain standard records of structural fields; S103: Based on the standard record of the structure field, the record fields are matched with the experimental design requirements, the fields that meet the requirements are compared and a mapping relationship is established, and the records that meet the experimental requirements are screened in combination with the dynamic follow-up management node evaluation to obtain the experimental associated health data; The specific steps of S2 are: S201: Based on the test-related health data, each field is compared item by item according to the preset inclusion criteria, and field features consistent with the standard content are recorded to generate field matching statistics; S202: Based on the field matching statistical results, sort the records that meet the inclusion criteria by reception time, calculate the reception time interval between adjacent records, obtain the time difference between every two records, and generate the average record time interval; S203: Based on the mean of the recording time intervals, the synchronization rhythm of the data batch updates is adjusted, the data update frequency is calculated, and the batches are allocated according to the update frequency. The patient records that meet the inclusion criteria are included in the update queue to generate the AI-screened qualified patient group; The calculation formula for the data update frequency is specifically: ； in, Represents the data update frequency, Represents the receiving time of the i-th record, Represents the difference between the i-th record and the previous record, Represents the total number of records, Represents the time difference between adjacent records, Represents the absolute value of the difference from the previous record.

2. The method for clinical trial patient screening and dynamic follow-up based on artificial intelligence according to claim 1, characterized in that: The associated health data specifically includes admission information, examination records, medication data, and imaging data. The group of patients qualified in the initial screening by artificial intelligence includes screening condition matching records, time interval data, and data update frequency. The dynamic follow-up management monitoring list includes medical unit identification, number of patients, and distribution difference marks. The health fluctuation trend of the dynamically monitored population specifically refers to time series, health difference, and fluctuation labels. The artificial intelligence-driven patient dynamic management plan includes follow-up time nodes, sequence adjustment details, and update items.

3. The method for clinical trial patient screening and dynamic follow-up based on artificial intelligence according to claim 1, characterized in that: The specific steps of S3 are: S301: Based on the AI-preliminary screening of qualified patients, extract the name of the medical institution and the corresponding number of patients associated with each record, group the records by medical institution, count the number of patients in each institution, and calculate the difference in the number of patients between institutions to generate a statistical table of the difference in the number of patients per institution; S302: Based on the unit patient quantity difference statistics table, partition the quantity difference ranges between units, classify units with deviations in the difference as high-priority monitoring areas, mark the units as areas with high data update frequency, extract the corresponding patient records, and generate a unit difference classification marking table; S303: Based on the unit difference classification mark table, the patient records in the high priority area are included in the dynamic follow-up management focus range, a dynamic follow-up management task schedule is generated, and the monitoring list is updated synchronously to obtain a dynamic follow-up management monitoring list.

4. The method for clinical trial patient screening and dynamic follow-up based on artificial intelligence according to claim 3, characterized in that: The specific calculation formula for the quantity difference between the units is: ； in, Represents the quantity difference between units, Representative The number of patients in each medical unit, Representative The number of patients in each medical unit, Representative Unit The adjustment value of Representative Unit The adjustment value of .

5. The method for clinical trial patient screening and dynamic follow-up based on artificial intelligence according to claim 1, characterized in that: The specific calculation formula for the numerical difference between adjacent records is: ； in, Represents the numerical difference between adjacent records. Representative Daily data record value, Represents the data record value of the previous day, Representative The daily weight value, Representative The data weight value of the day.

6. The method for clinical trial patient screening and dynamic follow-up based on artificial intelligence according to claim 1, characterized in that: The specific steps of S5 are: S501: Based on the health fluctuation trend of the dynamically monitored population, extract the patient number for which the follow-up plan needs to be adjusted, call the individual patient's scheduling record, search and match the associated follow-up nodes, obtain the time interval of each node, and generate the patient follow-up node time interval; S502: Based on the patient follow-up node time interval, compare the node schedule with the preset time, identify the node content with time overlap or inconsistency, record and mark the nodes that need to be adjusted, and generate a time node adjustment mark table; S503: Based on the time node adjustment mark table, adjust the trigger order of the nodes with overlap and inconsistency, set the newly added task node identifier, synchronize all updated information to the patient's individual plan table, and generate an artificial intelligence-driven patient dynamic management plan.

Images