Clinical data cleaning method, system and equipment based on large language model and medium

Abstract

The invention discloses a clinical data cleaning method, system and equipment based on a large language model and a medium, and relates to the technical field of clinical data processing, and the method comprises the following steps: analyzing clinical test data to obtain an ideal data relation graph; obtaining an actual data relation graph based on clinical experience data; performing difference analysis to obtain a difference fault-tolerant interval; establishing a data cleaning method in the large language model, and cleaning the clinical data; the method and device are used for solving the problems that in the prior art, abnormal data are directly corrected or complemented after the abnormal data are judged during data cleaning of clinical data, the reasonability of the abnormal data cannot be judged from the multi-dimensional perspective of the clinical medical field, the reasonably existing data are corrected, and the data recorded after correction are still inconsistent with the reality.

Description

Technical Field

[0001] The present invention relates to the technical field of clinical data processing, and in particular to a clinical data cleaning method, system, device and medium based on a large language model. Background Art

[0002] Clinical refers to direct contact with patients and actual observation of them; disciplines such as internal medicine, surgery, obstetrics and gynecology, and pediatrics belong to clinical medicine; in terms of the division of labor within the hospital, it is divided into clinical, medical technology, administration, and technical services departments; the clinical department is the main body of the hospital, and it is directly responsible for the admission, diagnosis, and treatment of patients; clinical personnel include doctors and nurses who are directly involved in the treatment and care of patients; the medical technology department is what was called the "non-clinical" department in the past, which also has doctors and nurses, but does not directly participate in the treatment and care of patients, but only provides direct services for clinical diagnosis and treatment.

[0003] Existing improvements in clinical data cleaning usually supplement and enter missing data in clinical data. As for abnormal judgment of abnormal data during data cleaning, the abnormal data is usually directly corrected after the abnormal data is judged, or the incomplete data is completed. This method of processing abnormal data cannot combine the relationship between various types of data in clinical medical data, so there will be a problem of inaccurate correction of abnormal data. For example, in a Chinese patent with application publication number CN118039048A, a clinical test data processing method is disclosed. This scheme is to improve the screening of abnormal data and correct the abnormal data after the abnormal data is screened. Although this improved method can check the abnormal data in the data, it cannot judge the rationality of the abnormal data from a multi-dimensional perspective in the clinical medical field, resulting in the correction of reasonable existing data, resulting in the problem that the recorded data after correction is still inconsistent with the actual situation. In view of this, it is necessary to improve the existing data cleaning method for clinical medical data. Summary of the invention

[0004] The present invention aims to solve one of the technical problems in the prior art to at least a certain extent. By proposing a clinical data cleaning method, system, device and medium based on a large language model, it is used to solve the problem of abnormal judgment of abnormal data during data cleaning in the prior art. Usually, after the abnormal data is judged, the abnormal data is directly corrected, or the incomplete data is completed. This method of processing abnormal data cannot combine the relationship between various types of data in clinical medical data, and therefore there is a problem of inaccurate correction of abnormal data.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a clinical data cleaning method based on a large language model, comprising:

[0006] Obtain clinical trial data of medical devices in clinical data, analyze the clinical trial data and obtain an ideal data relationship diagram; obtain clinical experience data of medical devices in clinical data based on clinical medical data, analyze the clinical experience data and obtain an actual data relationship diagram;

[0007] Perform difference analysis on the ideal data relationship diagram and the actual data relationship diagram to obtain the difference tolerance interval;

[0008] The difference tolerance interval is placed in the large language model and a data cleaning method is established in the large language model, and the clinical data is cleaned based on the large language model.

[0009] Furthermore, clinical trial data of medical devices in clinical data are obtained, and the clinical trial data are analyzed to obtain an ideal data relationship diagram including:

[0010] For any medical device A in clinical data α , obtain medical device A α When conducting clinical trials, the clinical trial data including the data of personnel involved in the safety data of the clinical trial data is recorded as personnel safety data. Based on the personnel safety data, the relationship analysis method is used to obtain the A α Ideal data relationship diagram;

[0011] The relationship analysis method includes: recording the adverse events recorded in the personnel safety data as personnel adverse events 1 to personnel adverse events K in sequence; for any personnel adverse event K1, recording the evaluation levels used to judge the personnel adverse events as level 1 evaluation level to level Z evaluation level in sequence; obtaining the level of the personnel adverse event K1 being evaluated;

[0012] A plane rectangular coordinate system is established, recorded as the personnel relationship analysis coordinate system, wherein the unit of the X-axis of the personnel relationship analysis coordinate system is fixed to the judgment level of personnel adverse events, and the unit of the Y-axis of the personnel relationship analysis coordinate system is the personnel health level, the equipment damage level or the medication level.

[0013] Furthermore, obtaining clinical trial data of medical devices in clinical data, analyzing the clinical trial data and obtaining an ideal data relationship diagram also includes:

[0014] When the unit of the Y-axis of the personnel relationship coordinate system is the personnel health level, the health data of all the test participants in the personnel adverse event K1 are obtained, and the health level of each test participant is obtained based on the health level classification, and the X-axis of the personnel relationship coordinate system is punctuated based on the level of the personnel adverse event K1, and the punctuation is recorded as the event level point; the Y-axis of the personnel relationship coordinate system is punctuated based on the health level of each test participant, and the punctuation is recorded as health level point 1 to health level point D, where D is the number of all test participants in the personnel adverse event K1; lines are connected from health level point 1 to health level point D to the event level points respectively, and all the line segments obtained after the connection are fitted into one line segment, which is recorded as the health relationship line segment;

[0015] When the unit of the Y-axis of the personnel relationship coordinate system is the device damage level, the judgment standard of the medical device damage level is obtained based on big data, and the medical device A in the personnel adverse event K1 is obtained based on the judgment standard. α Damage level; based on the medical device A in the personnel adverse event K1 α The damage level is marked on the Y axis of the personnel relationship coordinate system, and the line segment connecting the marked point and the event level point is recorded as the device relationship line segment;

[0016] When the unit of the Y-axis of the personnel relationship coordinate system is the medication grade, the grade classification of drugs is obtained based on big data, and the grades of all drugs used in the personnel adverse event K1 are obtained based on the drug grade classification, and the grades of all drugs used in the personnel adverse event K1 are punctuated on the Y-axis of the personnel relationship coordinate system, and are recorded as drug grade point YP1 to drug grade point YPS, respectively, where S is the number of drugs used in the personnel adverse event K1; lines are connected from drug grade point YP1 to drug grade point YPS to the event grade points, and all line segments obtained after the connection are fitted into a line segment, which is recorded as the drug relationship line segment;

[0017] Create a triangle, record it as an expanded triangle, where the three vertices of the expanded triangle are record as health, equipment and medicine respectively; record the lengths of the health relationship line segment, the equipment relationship segment and the medicine relationship segment as L1, L2 and L3 respectively; draw line segments from the center of the expanded triangle to the three vertices, the lengths of the line segments are L1, L2 and L3 respectively, connect the non-overlapping endpoints of the three line segments, and record the resulting triangle as the ideal data relationship diagram;

[0018] Obtain the ideal data relationship diagram of all medical devices based on relationship analysis methods.

[0019] Furthermore, clinical experience data of medical devices in clinical data are obtained based on clinical medical data, and the clinical experience data are analyzed to obtain actual data relationship diagrams including:

[0020] Based on the analysis method of clinical trial data, clinical experience data of medical devices are obtained, and the clinical experience data are analyzed based on the relationship analysis method. The triangles obtained when analyzing each medical device in the relationship analysis method are recorded as the actual data relationship diagram of the medical device.

[0021] Furthermore, by performing a difference analysis on the ideal data relationship diagram and the actual data relationship diagram, the difference tolerance intervals include:

[0022] A plane rectangular coordinate system is established, which is recorded as the difference analysis coordinate system, where the units of the X-axis and the Y-axis of the difference analysis coordinate system are both lengths; for any medical device A β , medical device A β The ideal data relationship diagram and the actual data relationship diagram are both placed in the difference analysis coordinate system, and when placing them, the centers of the expansion triangles corresponding to the ideal data relationship diagram and the actual data relationship diagram coincide with the origin of the difference analysis coordinate system;

[0023] The area where the ideal data relationship diagram overlaps with the actual data relationship diagram is recorded as the common area, the area only in the ideal data relationship diagram is recorded as the ideal area; the area only in the actual data relationship diagram is recorded as the actual area, and the difference analysis algorithm is used to obtain the medical device A β The difference valley value, the difference analysis algorithm includes: Among them, G is the difference valley value, γ is the common coefficient, is the area of ​​the common area, s1 is the area of ​​the ideal area, s2 is the area of ​​the actual area, s3 is the area of ​​the ideal data relationship diagram, and s4 is the area of ​​the actual data relationship diagram; the interval [difference valley, 1) is recorded as medical device A β The difference tolerance interval.

[0024] Furthermore, by performing a difference analysis on the ideal data relationship diagram and the actual data relationship diagram, the difference tolerance interval also includes:

[0025] Based on medical device A β The analysis method obtains the ideal data relationship diagram based on all medical devices and the actual data relationship diagram to obtain the difference tolerance interval of all medical devices.

[0026] Furthermore, the difference tolerance interval is put into the large language model and a data cleaning method is established in the large language model. The cleaning of clinical data based on the large language model includes:

[0027] storing the relationship analysis method and the method for obtaining the difference tolerance interval in a large language model, using the large language model to record the clinical medical data, and after recording the clinical medical data, using the data cleaning method to clean the clinical medical data recorded in the large language model;

[0028] The data cleaning method includes: for any clinical medical data Q recorded in a large language model, the medical device in the clinical medical data Q is recorded as the device to be analyzed, and the analysis is performed based on the relationship analysis method of the clinical medical data, and the triangle obtained during the analysis is recorded as the cleaning data relationship diagram of the device to be analyzed; the cleaning data relationship diagram and the ideal data relationship diagram of the device to be analyzed are placed in a difference analysis coordinate system, and the difference analysis algorithm is used to obtain the difference valley value corresponding to the cleaning data relationship diagram and the ideal data relationship diagram of the device to be analyzed, which is recorded as the actual difference value; when the actual difference value is within the difference tolerance interval of the device to be analyzed, the clinical medical data Q is recorded as data that does not need to be cleaned; when the actual difference value is outside the difference tolerance interval of the device to be analyzed, the clinical medical data Q is deleted.

[0029] In a second aspect, the present invention also provides a clinical data cleaning system based on a large language model, including a data relationship analysis module, a fault tolerance analysis training module and a data cleaning module;

[0030] The data relationship analysis module is used to obtain clinical trial data of medical devices in clinical data, analyze the clinical trial data and obtain an ideal data relationship diagram; obtain clinical experience data of medical devices in clinical data based on clinical medical data, analyze the clinical experience data and obtain an actual data relationship diagram;

[0031] The fault tolerance analysis training module is used to perform difference analysis on the ideal data relationship diagram and the actual data relationship diagram to obtain the difference fault tolerance interval;

[0032] The data cleaning module is used to put the difference tolerance interval into the large language model and establish a data cleaning method in the large language model, and clean the clinical data based on the large language model.

[0033] In a third aspect, the present invention provides a storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps in the above method are performed.

[0034] In a fourth aspect, the present invention provides an electronic device, comprising a processor and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps in the above method are performed.

[0035] Beneficial effects of the present invention: The present invention first obtains an ideal data relationship diagram for clinical trial data in clinical data; and obtains an actual data relationship diagram based on clinical experience data analysis. This has the advantage that, by obtaining the ideal data relationship diagram and the actual data relationship diagram, it is helpful to obtain the difference between the clinical data of the medical device when it is ideally used and when it is actually used, so that in the subsequent data cleaning, the relationship between various types of data in the clinical medical data can be combined for cleaning, so as to prevent the correction result from not conforming to the actual situation when the abnormal data is corrected;

[0036] The present invention also performs a difference analysis on the ideal data relationship diagram and the actual data relationship diagram to obtain a difference tolerance interval, and finally puts the difference tolerance interval into a large language model and establishes a data cleaning method in the large language model, and cleans the clinical data based on the large language model. The advantage of this is that by performing data cleaning based on the difference tolerance interval and in combination with the large language model, it is possible to accurately delete abnormal data when analyzing a large amount of clinical data based on the large language model, and use the difference tolerance interval before each deletion to determine whether the abnormal data is reasonable, thereby ensuring that each data cleaning can fit the actual situation of the clinical data. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a functional block diagram of the system of the present invention;

[0038] Figure 2 is a flow chart of the steps of the method of the present invention;

[0039] Figure 3 A schematic diagram of a personnel relationship analysis coordinate system of the present invention;

[0040] Figure 4 A schematic diagram of obtaining an ideal data relationship diagram of the present invention;

[0041] Figure 5 It is a schematic structural diagram of the electronic device of the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] Example 1, please refer to Figure 1 As shown, in the first aspect, the present application provides a clinical data cleaning system based on a large language model, including a data relationship analysis module, a fault tolerance analysis training module and a data cleaning module;

[0044] The data relationship analysis module is used to obtain clinical trial data of medical devices in clinical data, analyze the clinical trial data and obtain an ideal data relationship diagram; obtain clinical experience data of medical devices in clinical data based on clinical medical data, analyze the clinical experience data and obtain an actual data relationship diagram.

[0045] The data analysis module is configured with an ideal relationship analysis unit and an actual relationship analysis unit; the ideal relationship analysis unit is configured with an ideal relationship graph acquisition strategy;

[0046] The ideal relationship diagram acquisition strategy includes: for any medical device A in the clinical data α , obtain medical device A α When conducting clinical trials, the clinical trial data including the data of personnel involved in the safety data of the clinical trial data is recorded as personnel safety data. Based on the personnel safety data, the relationship analysis method is used to obtain the A α Ideal data relationship diagram;

[0047] In the specific implementation process, for example, during a data analysis, medical device A α If it is a ventilator, the data from the clinical trial when the ventilator was not put into use should be recorded as clinical trial data, and the personnel safety data should include the data of the personnel operating and wearing the ventilator;

[0048] The relationship analysis method includes: recording the adverse events recorded in the personnel safety data as personnel adverse events 1 to personnel adverse events K in sequence; for any personnel adverse event K1, recording the evaluation levels used to judge the personnel adverse events as level 1 evaluation level to level Z evaluation level in sequence; obtaining the level of the personnel adverse event K1 being evaluated;

[0049] In the specific implementation process, in this embodiment, the evaluation level of the adverse events of the personnel is set to level one adverse event to level five adverse events, wherein the level one adverse event is that the event result slightly interferes with the patient's life, but no special treatment is required, the level two adverse event is that the event result has a certain impact on the patient's life, and the drug dosage or nutritional support needs to be adjusted, the level three adverse event is that the event result has a greater impact on the patient's quality of life, such as severe diarrhea or moderate anemia, the level four adverse event is that the event result threatens the patient's life, and emergency treatment measures should be taken, and the level five adverse event is that the event result may cause the patient's death, such as severe heart problems or respiratory failure; when the specific implementation analysis is carried out, the evaluation level of the adverse event can be specifically set according to the actual medical device and the actual usage scenario; for example, when analyzing the adverse events in the personnel safety data of the ventilator, the personnel adverse event K1 is obtained as unstable oxygen flow, and unstable oxygen flow may cause respiratory failure or death of the patient, so the evaluation level of the personnel adverse event k1 should be recorded as a level five adverse event;

[0050] A plane rectangular coordinate system is established, recorded as the personnel relationship analysis coordinate system, wherein the unit of the X-axis of the personnel relationship analysis coordinate system is fixed to the judgment level of personnel adverse events, and the unit of the Y-axis of the personnel relationship analysis coordinate system is the personnel health level, the equipment damage level or the medication level.

[0051] In the specific implementation process, in the personnel relationship analysis coordinate system of this embodiment, for any one of the personnel health level, equipment damage level or medication level, the level number is set as the vertical coordinate of the point. For example, when analyzing the personnel health level, the coordinate corresponding to the second level of health is (0, 2); please refer to Figure 3 As shown, the coordinate system where D1 is located is the personnel relationship analysis coordinate system when the ordinate is the personnel health level, the coordinate system where D2 is located is the personnel relationship analysis coordinate system when the ordinate is the device damage level, and the coordinate system where D3 is located is the personnel relationship analysis when the ordinate is the medication level; r1 to r5 are coordinate points corresponding to level 1 health to level 5 health, q1 to q3 are coordinate points corresponding to level 1 damage to level 3 damage, y1 to y3 are coordinate points corresponding to level 1 drugs to level 3 drugs, and b1 to b5 are coordinate points corresponding to level 1 adverse events to level 5 adverse events;

[0052] When the unit of the Y-axis of the personnel relationship coordinate system is the personnel health level, the health data of all the test participants in the personnel adverse event K1 are obtained, and the health level of each test participant is obtained based on the health level classification, and the X-axis of the personnel relationship coordinate system is punctuated based on the level of the personnel adverse event K1, and the punctuation is recorded as the event level point; the Y-axis of the personnel relationship coordinate system is punctuated based on the health level of each test participant, and the punctuation is recorded as health level point 1 to health level point D, where D is the number of all test participants in the personnel adverse event K1; lines are connected from health level point 1 to health level point D to the event level points respectively, and all the line segments obtained after the connection are fitted into one line segment, which is recorded as the health relationship line segment;

[0053] In the specific implementation process, in this embodiment, the health level division is set to five levels, namely, level one health, normal physical indicators, full of energy, and no discomfort symptoms; level two health, occasional slight discomfort, but overall good health; level three health, often feeling tired, occasional minor illnesses and pains, and low immunity; level four health, obvious discomfort symptoms or confirmed chronic diseases; level five health, suffering from serious diseases or multiple chronic diseases, and poor health; in specific analysis, the health level division can be modified according to the actual health status of the personnel; for example, when analyzing the personnel adverse event K1 in the personnel safety data of the ventilator, the personnel adverse event K1 is unstable oxygen flow, and unstable oxygen flow is a level five adverse event. The number of test personnel in the personnel adverse event K1 is 4. Through the above health level division, the health level of each person is obtained as level one health, level two health, level one health and level three health, then the health level points 1 to 4 are (0, 1), (0, 2), (0, 1) and (0, 3), respectively. Please refer to Figure 3 As shown, the line segments connected by r1 and b5, r2 and b5, r1 and b5, and r3 and b5 are the lines connecting all health level points and event level points respectively. The line connecting r2 and b5 is obtained by fitting as the health relationship line segment;

[0054] When the unit of the Y-axis of the personnel relationship coordinate system is the device damage level, the judgment standard of the medical device damage level is obtained based on big data, and the medical device A in the personnel adverse event K1 is obtained based on the judgment standard. α Damage level; based on the medical device A in the personnel adverse event K1 α The damage level is marked on the Y axis of the personnel relationship coordinate system, and the line segment connecting the marked point and the event level point is recorded as the device relationship line segment;

[0055] In the specific implementation process, in this embodiment, the judgment standard of the damage level of medical devices is set to three levels, namely: first-level damage, which has a small impact on the performance of the medical device and can be solved by repairing or replacing parts; second-level damage, which has a large impact on the performance of the medical device, but it can still be used and needs regular maintenance; third-level damage: it has a great impact on the performance of the medical device, seriously affects the safety of use, and needs to be scrapped immediately; in the specific analysis, the judgment standard of the damage level of the medical device can be modified according to the medical device analyzed in the actual situation and the results obtained through big data. For example, when analyzing the adverse event K1 of the personnel safety data of the ventilator, the adverse event K1 is unstable oxygen flow, and the unstable oxygen flow is a fifth-level adverse event. The damage level of the ventilator is second-level damage, please refer to Figure 3 As shown, the line connecting q2 and b5 is the instrument relationship line;

[0056] When the unit of the Y-axis of the personnel relationship coordinate system is the medication grade, the grade classification of drugs is obtained based on big data, and the grades of all drugs used in the personnel adverse event K1 are obtained based on the drug grade classification, and the grades of all drugs used in the personnel adverse event K1 are punctuated on the Y-axis of the personnel relationship coordinate system, and are recorded as drug grade point YP1 to drug grade point YPS, respectively, where S is the number of drugs used in the personnel adverse event K1; lines are connected from drug grade point YP1 to drug grade point YPS to the event grade points, and all line segments obtained after the connection are fitted into a line segment, which is recorded as the drug relationship line segment;

[0057] In the specific implementation process, in this embodiment, the classification of drugs is specifically as follows: Class I drugs are drugs with high toxicity, high risk, easy to abuse or easy to form dependence; Class II drugs are drugs with low toxicity, low risk, not easy to abuse or not easy to form dependence; Class III drugs have low toxicity and low risk, and are not easy to abuse or form dependence; in specific analysis, the classification of drugs can be modified according to the drug taking situation of the test personnel and the specific classification of drugs. For example, when analyzing the personnel adverse event K1 in the personnel safety data of the ventilator, the personnel adverse event K1 is unstable oxygen flow, and unstable oxygen flow is a fifth-level adverse event. The level of all drugs used in the personnel adverse event K1 is third-level drugs, then the drug relationship line segment is Figure 3 The line connecting y3 and b5;

[0058] Create a triangle, record it as an expanded triangle, where the three vertices of the expanded triangle are record as health, equipment and medicine respectively; record the lengths of the health relationship line segment, the equipment relationship segment and the medicine relationship segment as L1, L2 and L3 respectively; draw line segments from the center of the expanded triangle to the three vertices, the lengths of the line segments are L1, L2 and L3 respectively, connect the non-overlapping endpoints of the three line segments, and record the resulting triangle as the ideal data relationship diagram;

[0059] In the specific implementation process, please refer to Figure 4 As shown, points SS1 to SS3 are health, equipment and medicine respectively, the lengths of line segments J1 to J3 are L1, L2 and L3 respectively, and the triangle formed by points SS4, SS5 and SS6 is an ideal data relationship diagram;

[0060] Obtain the ideal data relationship diagram of all medical devices based on relationship analysis methods.

[0061] The actual relationship analysis unit is configured with an actual relationship graph acquisition strategy, which includes: acquiring clinical experience data of medical devices based on an analysis method of clinical trial data, and analyzing the clinical experience data based on a relationship analysis method, and recording the triangles obtained when analyzing each medical device in the relationship analysis method as the actual data relationship graph of the medical device.

[0062] The fault-tolerant analysis training module is used to perform difference analysis on the ideal data relationship diagram and the actual data relationship diagram to obtain the difference fault-tolerant interval.

[0063] The fault-tolerant analysis training module includes a difference analysis unit, and the difference analysis unit is configured with a difference fault-tolerant analysis strategy. The difference fault-tolerant analysis strategy includes:

[0064] A plane rectangular coordinate system is established, which is recorded as the difference analysis coordinate system, where the units of the X-axis and the Y-axis of the difference analysis coordinate system are both lengths; for any medical device A β , medical device A β The ideal data relationship diagram and the actual data relationship diagram are both placed in the difference analysis coordinate system, and when placing them, the centers of the expansion triangles corresponding to the ideal data relationship diagram and the actual data relationship diagram coincide with the origin of the difference analysis coordinate system;

[0065] The area where the ideal data relationship diagram overlaps with the actual data relationship diagram is recorded as the common area, the area only in the ideal data relationship diagram is recorded as the ideal area; the area only in the actual data relationship diagram is recorded as the actual area, and the difference analysis algorithm is used to obtain the medical device A β The difference valley value, the difference analysis algorithm includes: Among them, G is the difference valley value, γ is the common coefficient, is the area of ​​the common area, s1 is the area of ​​the ideal area, s2 is the area of ​​the actual area, s3 is the area of ​​the ideal data relationship diagram, and s4 is the area of ​​the actual data relationship diagram; the interval [difference valley, 1) is recorded as medical device A β The difference tolerance interval.

[0066] In the specific implementation process, in this embodiment, since the number of overlapping graphs is 2, the value of γ is 2. For example, in a medical device A β When the actual analysis is performed, the area of ​​the common area is 10cm 2 , s1 is 2cm 2 , s2 is 5cm 2 , s3 is 12cm 2 , s4 is 15cm 2 , then by calculation, the difference valley value is about 0.239, and the interval [0.239, 1) is the interval of medical device A β In the difference analysis algorithm, when the ideal data relationship diagram of the medical device to be analyzed completely overlaps with the actual data relationship diagram, the difference valley value obtained by calculation is 1, indicating that there can be no deviation between the actual application data of the medical device and the data during the test;

[0067] Based on medical device A β The analysis method obtains the ideal data relationship diagram based on all medical devices and the actual data relationship diagram to obtain the difference tolerance interval of all medical devices.

[0068] The data cleaning module is used to put the difference tolerance interval into the large language model and establish a data cleaning method in the large language model, and clean the clinical data based on the large language model.

[0069] The data cleaning module includes a data cleaning unit, the data cleaning unit is configured with a clinical data cleaning strategy, the clinical data cleaning strategy includes: storing a relationship analysis method and a method for obtaining a difference tolerance interval in a large language model, using the large language model to record clinical medical data, and after recording the clinical medical data, using the data cleaning method to clean the clinical medical data recorded in the large language model;

[0070] In the specific implementation process, the use of large language models helps to comprehensively clean the data when facing massive amounts of clinical data, thus preventing loopholes in data cleaning caused by excessive amounts of data;

[0071] The data cleaning method includes: for any clinical medical data Q recorded in a large language model, the medical device in the clinical medical data Q is recorded as the device to be analyzed, and the analysis is performed based on the relationship analysis method of the clinical medical data, and the triangle obtained during the analysis is recorded as the cleaning data relationship diagram of the device to be analyzed; the cleaning data relationship diagram and the ideal data relationship diagram of the device to be analyzed are placed in a difference analysis coordinate system, and the difference analysis algorithm is used to obtain the difference valley value corresponding to the cleaning data relationship diagram and the ideal data relationship diagram of the device to be analyzed, which is recorded as the actual difference value; when the actual difference value is within the difference tolerance interval of the device to be analyzed, the clinical medical data Q is recorded as data that does not need to be cleaned; when the actual difference value is outside the difference tolerance interval of the device to be analyzed, the clinical medical data Q is deleted.

[0072] In the specific implementation process, when the actual difference value is within the difference tolerance interval of the device to be analyzed, it means that although the data corresponding to the actual difference value may be different from the existing data, the difference is within a controllable range. Therefore, there is no need to clean the clinical medical data whose actual difference value is within the difference tolerance interval of the device to be analyzed; and for the data whose actual difference value is outside the difference tolerance interval of the device to be analyzed, it means that there is a large difference between the data and the actual standard data, and the existing data does not exist in the data that matches the data, so the data should be cleaned, that is, the data should be deleted.

[0073] Example 2, please refer to Figure 2 As shown, in a second aspect, the present invention provides a clinical data cleaning method based on a large language model, comprising:

[0074] Step S1, obtaining clinical trial data of medical devices in clinical data, analyzing the clinical trial data and obtaining an ideal data relationship diagram; obtaining clinical experience data of medical devices in clinical data based on clinical medical data, analyzing the clinical experience data and obtaining an actual data relationship diagram.

[0075] Step S1 includes the following sub-steps:

[0076] Step S101: for any medical device A in the clinical data α , obtain medical device A α When conducting clinical trials, the clinical trial data including the data of personnel involved in the safety data of the clinical trial data is recorded as personnel safety data. Based on the personnel safety data, the relationship analysis method is used to obtain the A α Ideal data relationship diagram;

[0077] Step S102, the relationship analysis method includes: recording the adverse events recorded in the personnel safety data as personnel adverse events 1 to personnel adverse events K in sequence; for any personnel adverse event K1, recording the evaluation levels used to judge the personnel adverse events as level 1 evaluation level to level Z evaluation level in sequence; and obtaining the level of the personnel adverse event K1 to be evaluated;

[0078] Step S103, establish a plane rectangular coordinate system, recorded as the personnel relationship analysis coordinate system, wherein the unit of the X-axis of the personnel relationship analysis coordinate system is fixed to the judgment level of personnel adverse events, and the unit of the Y-axis of the personnel relationship analysis coordinate system is the personnel health level, equipment damage level or medication level.

[0079] Step S104, when the unit of the Y-axis of the personnel relationship coordinate system is the personnel health level, the health data of all the test personnel participating in the personnel adverse event K1 is obtained, and the health level of each test personnel is obtained based on the health level classification, and the X-axis of the personnel relationship coordinate system is punctuated based on the level of the personnel adverse event K1, and the punctuation is recorded as the event level point; the Y-axis of the personnel relationship coordinate system is punctuated based on the health level of each test personnel, and the punctuation is recorded as health level point 1 to health level point D, where D is the number of all the test personnel participating in the personnel adverse event K1; lines are connected from health level point 1 to health level point D to the event level point respectively, and all the line segments obtained after the connection are fitted into one line segment, which is recorded as the health relationship line segment;

[0080] Step S105: when the unit of the Y axis of the personnel relationship coordinate system is the device damage level, the judgment standard of the medical device damage level is obtained based on the big data, and the medical device A in the personnel adverse event K1 is obtained based on the judgment standard. α Damage level; based on the medical device A in the personnel adverse event K1 α The damage level is marked on the Y axis of the personnel relationship coordinate system, and the line segment connecting the marked point and the event level point is recorded as the device relationship line segment;

[0081] Step S106, when the unit of the Y-axis of the personnel relationship coordinate system is the medication level, the medication level classification is obtained based on the big data, the levels of all medications used in the personnel adverse event K1 are obtained based on the medication level classification, and the levels of all medications used in the personnel adverse event K1 are marked in the Y-axis of the personnel relationship coordinate system, and are respectively recorded as medication level point YP1 to medication level point YPS, where S is the number of medications used in the personnel adverse event K1; lines are respectively connected from medication level point YP1 to medication level point YPS to the event level points, and all line segments obtained after the connection are fitted into one line segment, which is recorded as the medication relationship line segment;

[0082] Step S107, creating a triangle, recorded as an expanded triangle, wherein the three vertices of the expanded triangle are recorded as health, device, and drug respectively; the lengths of the health relationship line segment, the device relationship segment, and the drug relationship segment are recorded as L1, L2, and L3 respectively; drawing line segments from the center of the expanded triangle to the three vertices, the lengths of the line segments are L1, L2, and L3 respectively, connecting the non-overlapping endpoints of the three line segments, and recording the obtained triangle as an ideal data relationship diagram;

[0083] Step S108, obtaining an ideal data relationship diagram of all medical devices based on a relationship analysis method.

[0084] Step S109, based on the analysis method of clinical trial data, clinical experience data of medical devices are obtained, and the clinical experience data are analyzed based on the relationship analysis method, and the triangles obtained when analyzing each medical device in the relationship analysis method are recorded as the actual data relationship diagram of the medical device.

[0085] Step S2, performing difference analysis on the ideal data relationship diagram and the actual data relationship diagram to obtain the difference tolerance interval. Step S2 includes the following sub-steps: Step S201, establishing a plane rectangular coordinate system, recorded as the difference analysis coordinate system, wherein the units of the X-axis and the Y-axis of the difference analysis coordinate system are both lengths; for any medical device A β , medical device A β The ideal data relationship diagram and the actual data relationship diagram are both placed in the difference analysis coordinate system, and when placing them, the centers of the expansion triangles corresponding to the ideal data relationship diagram and the actual data relationship diagram coincide with the origin of the difference analysis coordinate system;

[0086] Step S202: record the area where the ideal data relationship diagram and the actual data relationship diagram overlap as the common area, record the area only in the ideal data relationship diagram as the ideal area, and record the area only in the actual data relationship diagram as the actual area, and use the difference analysis algorithm to obtain the medical device A β The difference valley value, the difference analysis algorithm includes: Among them, G is the difference valley value, γ is the common coefficient, is the area of ​​the common area, s1 is the area of ​​the ideal area, s2 is the area of ​​the actual area, s3 is the area of ​​the ideal data relationship diagram, and s4 is the area of ​​the actual data relationship diagram; the interval [difference valley, 1) is recorded as medical device A β The difference tolerance interval.

[0087] Step S203, based on the medical device A β The analysis method obtains the ideal data relationship diagram based on all medical devices and the actual data relationship diagram to obtain the difference tolerance interval of all medical devices.

[0088] Step S3, putting the difference tolerance interval into the large language model and establishing a data cleaning method in the large language model, and cleaning the clinical data based on the large language model. Step S3 includes the following sub-steps:

[0089] Step S301, storing the relationship analysis method and the method for obtaining the difference tolerance interval in a large language model, using the large language model to record clinical medical data, and after recording the clinical medical data, using a data cleaning method to clean the clinical medical data recorded in the large language model;

[0090] Step S302, the data cleaning method includes: for any clinical medical data Q recorded in the large language model, the medical device in the clinical medical data Q is recorded as the device to be analyzed, and the analysis is performed based on the relationship analysis method of the clinical medical data, and the triangle obtained during the analysis is recorded as the cleaning data relationship diagram of the device to be analyzed; the cleaning data relationship diagram and the ideal data relationship diagram of the device to be analyzed are placed in the difference analysis coordinate system, and the difference analysis algorithm is used to obtain the difference valley value corresponding to the cleaning data relationship diagram and the ideal data relationship diagram of the device to be analyzed, and recorded as the actual difference value; when the actual difference value is within the difference tolerance interval of the device to be analyzed, the clinical medical data Q is recorded as data that does not need to be cleaned; when the actual difference value is outside the difference tolerance interval of the device to be analyzed, the clinical medical data Q is deleted.

[0091] Example 3, third aspect, please refer to Figure 5 As shown, the present application provides an electronic device 40, including a processor 401 and a memory 402, the memory 402 stores computer-readable instructions, and when the computer-readable instructions are executed by the processor 401, the steps in the above method are executed. Through the above technical solution, the processor 401 and the memory 402 are interconnected and communicate with each other through a communication bus and / or other forms of connection mechanisms (not shown), and the memory 402 stores a computer program executable by the processor 401. When the electronic device is running, the processor 401 executes the computer program to execute the method in any optional implementation of the above embodiment to achieve the following functions: first, an ideal data relationship diagram is obtained for the clinical trial data in the clinical data; based on the clinical experience data analysis, an actual data relationship diagram is obtained, and a difference analysis is performed on the ideal data relationship diagram and the actual data relationship diagram to obtain a difference tolerance interval, and finally the difference tolerance interval is placed in a large language model and a data cleaning method is established in the large language model, and the clinical data is cleaned based on the large language model.

[0092] Embodiment 4, fourth aspect, the present application provides a storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps in the above method are executed. Through the above technical scheme, when the computer program is executed by the processor, the method in any optional implementation of the above embodiment is executed to achieve the following functions: first, an ideal data relationship diagram is obtained for the clinical trial data in the clinical data; based on the clinical experience data analysis, the actual data relationship diagram is obtained, and the ideal data relationship diagram and the actual data relationship diagram are analyzed for differences to obtain a difference tolerance interval, and finally the difference tolerance interval is put into a large language model and a data cleaning method is established in the large language model, and the clinical data is cleaned based on the large language model.

[0093] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0094] It should be understood by those skilled in the art that the embodiments of the present invention can be provided as methods, systems or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program codes. Among them, the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Read-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, disk or optical disk. These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0095] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

Claims

1. A clinical data cleaning method based on a large language model, characterized in that: include: Obtain clinical trial data of medical devices in clinical data, analyze clinical trial data and obtain an ideal data relationship diagram; Acquire clinical experience data of medical devices in clinical data based on clinical medical data, and analyze the clinical experience data to obtain an actual data relationship diagram; Perform difference analysis on the ideal data relationship diagram and the actual data relationship diagram to obtain the difference tolerance interval; The difference tolerance interval is placed in the large language model and a data cleaning method is established in the large language model, and the clinical data is cleaned based on the large language model.

2. The clinical data cleaning method based on a large language model according to claim 1, characterized in that: Obtain clinical trial data of medical devices in clinical data, analyze clinical trial data and obtain ideal data relationship diagram including: For any medical device A in clinical data α , obtain medical device A α When conducting clinical trials, the clinical trial data including the data of personnel involved in the safety data of the clinical trial data is recorded as personnel safety data. Based on the personnel safety data, the relationship analysis method is used to obtain the A α Ideal data relationship diagram; The relationship analysis method includes: recording the adverse events recorded in the personnel safety data as personnel adverse events 1 to personnel adverse events K in sequence; for any personnel adverse event K1, recording the evaluation levels used to judge the personnel adverse events as level 1 evaluation level to level Z evaluation level in sequence; obtaining the level of the personnel adverse event K1 being evaluated; A plane rectangular coordinate system is established, recorded as the personnel relationship analysis coordinate system, wherein the unit of the X-axis of the personnel relationship analysis coordinate system is fixed to the judgment level of personnel adverse events, and the unit of the Y-axis of the personnel relationship analysis coordinate system is the personnel health level, the equipment damage level or the medication level.

3. The clinical data cleaning method based on a large language model according to claim 2, characterized in that: Obtaining clinical trial data of medical devices in clinical data, analyzing clinical trial data and obtaining an ideal data relationship diagram also includes: When the unit of the Y-axis of the personnel relationship coordinate system is the personnel health level, the health data of all the test participants in the personnel adverse event K1 are obtained, and the health level of each test participant is obtained based on the health level classification, and the X-axis of the personnel relationship coordinate system is punctuated based on the level of the personnel adverse event K1, and the punctuation is recorded as the event level point; the Y-axis of the personnel relationship coordinate system is punctuated based on the health level of each test participant, and the punctuation is recorded as health level point 1 to health level point D, where D is the number of all test participants in the personnel adverse event K1; lines are connected from health level point 1 to health level point D to the event level points respectively, and all the line segments obtained after the connection are fitted into one line segment, which is recorded as the health relationship line segment; When the unit of the Y-axis of the personnel relationship coordinate system is the device damage level, the judgment standard of the medical device damage level is obtained based on big data, and the medical device A in the personnel adverse event K1 is obtained based on the judgment standard. α Damage level; based on the medical device A in the personnel adverse event K1 α The damage level is marked on the Y axis of the personnel relationship coordinate system, and the line segment connecting the marked point and the event level point is recorded as the device relationship line segment; When the unit of the Y-axis of the personnel relationship coordinate system is the medication grade, the grade classification of drugs is obtained based on big data, and the grades of all drugs used in the personnel adverse event K1 are obtained based on the drug grade classification, and the grades of all drugs used in the personnel adverse event K1 are punctuated on the Y-axis of the personnel relationship coordinate system, and are recorded as drug grade point YP1 to drug grade point YPS, respectively, where S is the number of drugs used in the personnel adverse event K1; lines are connected from drug grade point YP1 to drug grade point YPS to the event grade points, and all line segments obtained after the connection are fitted into a line segment, which is recorded as the drug relationship line segment; Create a triangle, record it as an expanded triangle, where the three vertices of the expanded triangle are record as health, equipment and medicine respectively; record the lengths of the health relationship line segment, the equipment relationship segment and the medicine relationship segment as L1, L2 and L3 respectively; draw line segments from the center of the expanded triangle to the three vertices, the lengths of the line segments are L1, L2 and L3 respectively, connect the non-overlapping endpoints of the three line segments, and record the resulting triangle as the ideal data relationship diagram; Obtain the ideal data relationship diagram of all medical devices based on relationship analysis methods.

4. The clinical data cleaning method based on a large language model according to claim 3, characterized in that: Based on clinical medical data, clinical experience data of medical devices in clinical data are obtained, and the actual data relationship diagram obtained by analyzing the clinical experience data includes: Based on the analysis method of clinical trial data, clinical experience data of medical devices are obtained, and the clinical experience data are analyzed based on the relationship analysis method. The triangles obtained when analyzing each medical device in the relationship analysis method are recorded as the actual data relationship diagram of the medical device.

5. The clinical data cleaning method based on a large language model according to claim 4, characterized in that: The difference analysis of the ideal data relationship diagram and the actual data relationship diagram is carried out, and the difference tolerance intervals are obtained, including: A plane rectangular coordinate system is established, which is recorded as the difference analysis coordinate system, where the units of the X-axis and the Y-axis of the difference analysis coordinate system are both lengths; for any medical device A β , medical device A β The ideal data relationship diagram and the actual data relationship diagram are both placed in the difference analysis coordinate system, and when placed, the centers of the expansion triangles corresponding to the ideal data relationship diagram and the actual data relationship diagram coincide with the origin of the difference analysis coordinate system; The area where the ideal data relationship diagram overlaps with the actual data relationship diagram is recorded as the common area, the area only in the ideal data relationship diagram is recorded as the ideal area; the area only in the actual data relationship diagram is recorded as the actual area, and the difference analysis algorithm is used to obtain the medical device A β The difference valley value, the difference analysis algorithm includes: Among them, G is the difference valley value, γ is the common coefficient, is the area of ​​the common area, s1 is the area of ​​the ideal area, s2 is the area of ​​the actual area, s3 is the area of ​​the ideal data relationship diagram, and s4 is the area of ​​the actual data relationship diagram; the interval [difference valley, 1) is recorded as medical device A β The difference tolerance interval.

6. The clinical data cleaning method based on a large language model according to claim 5, characterized in that: The difference analysis of the ideal data relationship diagram and the actual data relationship diagram shows that the difference tolerance interval also includes: Based on medical device A β The analysis method obtains the ideal data relationship diagram based on all medical devices and the actual data relationship diagram to obtain the difference tolerance interval of all medical devices.

7. The clinical data cleaning method based on a large language model according to claim 6, characterized in that: The difference tolerance interval is put into the large language model and a data cleaning method is established in the large language model. The cleaning of clinical data based on the large language model includes: storing the relationship analysis method and the method for obtaining the difference tolerance interval in a large language model, using the large language model to record the clinical medical data, and after recording the clinical medical data, using the data cleaning method to clean the clinical medical data recorded in the large language model; The data cleaning method includes: for any clinical medical data Q recorded in a large language model, the medical device in the clinical medical data Q is recorded as the device to be analyzed, and the analysis is performed based on the relationship analysis method of the clinical medical data, and the triangle obtained during the analysis is recorded as the cleaning data relationship diagram of the device to be analyzed; the cleaning data relationship diagram and the ideal data relationship diagram of the device to be analyzed are placed in a difference analysis coordinate system, and the difference analysis algorithm is used to obtain the difference valley value corresponding to the cleaning data relationship diagram and the ideal data relationship diagram of the device to be analyzed, which is recorded as the actual difference value; when the actual difference value is within the difference tolerance interval of the device to be analyzed, the clinical medical data Q is recorded as data that does not need to be cleaned; when the actual difference value is outside the difference tolerance interval of the device to be analyzed, the clinical medical data Q is deleted.

8. A clinical data cleaning system based on a large language model, used to implement the clinical data cleaning method based on a large language model according to any one of claims 1 to 7, characterized in that: It includes data relationship analysis module, fault-tolerant analysis training module and data cleaning module; The data relationship analysis module is used to obtain clinical trial data of medical devices in clinical data, analyze the clinical trial data and obtain an ideal data relationship diagram; obtain clinical experience data of medical devices in clinical data based on clinical medical data, analyze the clinical experience data and obtain an actual data relationship diagram; The fault tolerance analysis training module is used to perform difference analysis on the ideal data relationship diagram and the actual data relationship diagram to obtain the difference fault tolerance interval; The data cleaning module is used to put the difference tolerance interval into the large language model and establish a data cleaning method in the large language model, and clean the clinical data based on the large language model.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the steps in the method according to any one of claims 1 to 7 are executed.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps in the method according to any one of claims 1 to 7 are executed.

Images