Electronic information quality control system for intelligent clinical care

Abstract

The present invention relates to the field of medical information processing technology, and specifically to an electronic information quality control system for intelligent clinical nursing, comprising: a data acquisition module, used to obtain several types of nursing data for each patient. A data analysis module, used to obtain a first category of reference data; obtain a generalization interval of target data and a second category of target data for each patient according to the first category of reference data, and obtain a generalization interval of target data and a second category of target data for each patient again according to the second category of reference data; and abnormal data in the second category. A quality control module, used to judge the quality of target data of newly added patients according to the abnormal data in the second category and the generalization interval of abnormal data. The present invention timely and accurately judges the quality of target data of newly added patients, while further avoiding the influence of noise caused by individual differences among patients, and ultimately improves the quality of electronic information for clinical nursing.

Description

Technical Field

[0001] The present invention relates to the technical field of medical information processing, and in particular to an electronic information quality control system for intelligent clinical nursing. Background Art

[0002] In modern healthcare, with the widespread adoption of electronic medical records (EMR) and electronic health record (EHR) systems, information quality deficiencies have gradually become apparent. Complete and accurate patient information is crucial for intelligent clinical care, for example, in developing appropriate care plans. However, many current medical information systems face the following challenges in data recording and management: critical information such as patient medical history, diagnosis, and treatment records is often incomplete; data from multiple sources is poorly integrated, and patient information recorded by different medical professionals may contain conflicts or errors, leading to illogical clinical care or treatment processes. These issues directly impact the quality of electronic clinical care data, making it difficult to promptly identify low-quality data, resulting in unreliable and low-quality data. To fully leverage the potential of electronic clinical care data in intelligent applications, quality control and improvement of this directly recorded data are necessary. Summary of the Invention

[0003] In order to solve the above-mentioned problems of unreliable and low-quality clinical nursing electronic information data, the present invention provides an electronic information quality control system for intelligent clinical nursing.

[0004] The electronic information quality control system for intelligent clinical nursing of the present invention adopts the following technical solutions:

[0005] One embodiment of the present invention provides an electronic information quality control system for intelligent clinical nursing, which includes the following modules:

[0006] The data collection module is used to obtain several types of nursing data for each patient, including: disease details, medication data, and nursing monitoring results;

[0007] The data analysis module is used to record any type of nursing data as target data; nursing data other than target data are recorded as reference data and sample data in sequence. When the target data is disease details, the reference data and sample data are medication data and nursing monitoring results, respectively; when the target data is medication data, the reference data and sample data are disease details and nursing monitoring results, respectively; when the target data is nursing monitoring results, the reference data and sample data are disease details and medication data, respectively; the reference data of all patients are clustered to obtain the first category of the reference data;

[0008] Obtaining a generalization interval of target data for each patient and a second category of target data based on the first category of reference data, including: obtaining a plurality of most similar data of the target data of each patient in the first category of reference data, and obtaining a generalization interval of the target data of each patient based on differences between sample data corresponding to the plurality of most similar data; clustering the target data of all patients using similarities between the target data, the generalization interval, and whether the target data is within the generalization interval to obtain a plurality of second categories of target data;

[0009] For all nursing data of each patient, a generalization interval of all nursing data of each patient and a second category of each nursing data are obtained; the generalization interval of all nursing data includes the second category of the reference data;

[0010] Obtaining the generalization interval of the target data of each patient and the second category of the target data again based on the second category of the reference data; obtaining the offset error of the target data of each patient in the second category to which the target data of each patient belongs; and obtaining abnormal data in the second category based on the offset errors of the target data of all patients;

[0011] The quality control module determines the quality of the target data of the newly added patient based on the abnormal data in the second category and whether the target data of the newly added patient falls into the generalization interval of the abnormal data.

[0012] Preferably, clustering the reference data of all patients to obtain the first category of reference data comprises the following specific steps:

[0013] The similarity of the reference data of any two patients is obtained, denoted as x, and exp(-x) is denoted as the metric distance of the reference data of any two patients. The reference data of all patients are clustered using the K-Means clustering algorithm according to the metric distance to obtain several first categories. exp() represents an exponential function with a natural constant as the base.

[0014] Preferably, the step of obtaining a plurality of most similar data of the target data of each patient and obtaining a generalization interval of the target data of each patient based on the differences between the sample data corresponding to the plurality of most similar data comprises the following specific steps:

[0015] Each first category of the reference data includes target data corresponding to the reference data of each patient;

[0016] For all target data in any first category of the reference data, the target data of any patient is recorded as A. For each target data in the first category other than target data A, the similarity between target data A and each other target data is obtained, and the target data with the greatest similarity are taken as the most similar data to target data A.

[0017] For all sample data corresponding to the most similar data of target data A, the difference between all sample data is recorded as a;

[0018] The mean of the similarities between target data A and all the most similar data is recorded as wa; in the first category to which target data A belongs, the mean of the similarities between target data A and all other target data is obtained and recorded as the coordinate value of target data A, recorded as h;

[0019] The generalization interval of the target data A is recorded as [(1-a / wa)×h, (1+a / wa)×h].

[0020] Preferably, the specific steps for obtaining the differences between all sample data are as follows:

[0021] In all the sample data, the similarity between any two sample data is obtained, the mean of the similarities between all the sample data is recorded as b, exp(-b) is recorded as the difference between all the sample data, and exp() represents an exponential function with a natural constant as the base.

[0022] Preferably, clustering the target data of all patients using the similarity between the target data, the generalization interval, and whether the target data is within the generalization interval to obtain a plurality of second categories of target data includes the following specific steps:

[0023] The similarity of the target data of any two patients is recorded as y1; the intersection and union ratio of the generalization intervals of the target data of any two patients is recorded as y2;

[0024] Let Y = (1 + y2 × y3) × y1 be the corrected similarity of the target data of any two patients;

[0025] For any two patients’ target data Q1 and Q2, when the coordinate value of Q1 falls within the generalization interval of Q2 and the coordinate value of Q2 falls within the generalization interval of Q1, the parameter y3=1, otherwise the parameter y3=0;

[0026] exp(-Y) is used as the metric distance between the target data of any two patients, and then the target data of all patients are clustered using the metric distance to obtain several second categories of the target data; exp() represents an exponential function with a natural constant as the base.

[0027] Preferably, the steps of obtaining the generalization interval of the target data of each patient and the second category of the target data again based on the second category of the reference data; obtaining the offset error of the target data of each patient in the second category to which the target data of each patient belongs; and obtaining abnormal data in the second category based on the offset errors of the target data of all patients include the following specific steps:

[0028] H1: The second category of the reference data is used as the first category of the reference data; then, based on the first category of the reference data, the generalization interval of the target data and the second category of the target data of each patient are obtained. For all the nursing data of each patient, the generalization interval of all the nursing data of each patient and the second category of each nursing data are obtained; the generalization interval of all the nursing data includes the second category of the reference data;

[0029] H2: Repeat the steps in H1;

[0030] After each execution of the steps in H1, the offset error of the generalization interval of each target data in the second category to which the target data of each patient belongs is obtained; the set consisting of the offset errors of the generalization intervals of all target data in all second categories is recorded as the error distribution after each execution of H1; when the error distribution no longer changes, the execution of the steps in H1 is stopped; after stopping the execution of the steps in H1, the abnormal data in the second category is obtained according to the error distribution.

[0031] Preferably, the step of obtaining the offset error of the generalization interval of each target data includes the following specific steps:

[0032] For the target data in the second category, any generalization interval among the generalization intervals of all target data is recorded as B, and other generalization intervals other than B are recorded as reference intervals. The difference B1 between the left endpoint of B and the left endpoint of each reference interval is obtained, and the difference B2 between the right endpoint of B and the right endpoint of each reference interval is obtained. [B1, B2] is recorded as the error feature between each reference interval and B; the error features of all reference intervals and B are subjected to PCA dimensionality reduction, and the mean of all dimensionality reduction results is recorded as the offset error of the generalization interval B.

[0033] Preferably, the step of obtaining abnormal data in the second category according to the error distribution includes the following specific steps:

[0034] The final error distribution includes the offset error of the generalization interval of each target data; the LOF anomaly detection algorithm is used to obtain the abnormal offset error in the final error distribution, and the target data corresponding to the offset error is recorded as abnormal data.

[0035] Preferably, judging the quality of the target data of the newly added patient based on the abnormal data in the second category and whether the target data of the newly added patient falls within the generalization interval of the abnormal data includes the following specific steps:

[0036] Record the target data of the newly added patient as data F, and obtain the second category to which data F belongs among all the second categories of the target data;

[0037] For any abnormal data in the second category to which data F belongs, the similarity between data F and the abnormal data is recorded as the abnormal value of data F and the abnormal data. When the abnormal value falls into the generalization interval of the abnormal data, the abnormal data is recorded as the target abnormal data;

[0038] The target data other than all abnormal data in the second category to which data F belongs is recorded as normal data. The similarity between data F and normal data is recorded as the normal value of data F and normal data. When the normal value falls into the generalization interval of normal data, the normal data is recorded as the target normal data.

[0039] The quality of the target data of the newly added patients is judged based on the generalization interval between the target abnormal data and the target normal data.

[0040] Preferably, judging the quality of the target data of the newly added patient based on the generalization interval between the target abnormal data and the target normal data includes the following specific steps:

[0041] Obtain the intersection-and-union ratio of the generalization intervals of any pair of target abnormal data and target normal data, and record it as a normal indicator of data F. For all target abnormal data and target normal data, obtain all normal indicators of data F. When the maximum value of the normal indicator is less than the first preset threshold th1, the target data of the newly added patient is recorded as low-quality data, otherwise it is not recorded as low-quality data.

[0042] The beneficial effects of the technical solution of the present invention are:

[0043] In the first category of reference data, several most similar data of the target data of each patient are obtained, and the generalization interval of the target data of each patient is obtained based on the differences between the sample data corresponding to the several most similar data. The generalization interval is used to describe the generalization ability of the target data under the same or similar reference data (that is, in the same first category). The generalization ability indicates the degree to which the target data is affected by noise interference (for example, the degree to which it is affected by individual differences in patient weight, age, physiological condition, etc.).

[0044] Furthermore, the target data of all patients are clustered using the similarity between the target data, the generalization interval, and whether the target data is within the generalization interval to obtain several second categories of target data; the target data in the same second category not only have the same or similar characteristics, but also any target data is as far as possible within the generalization interval of another target data, ensuring that all target data in each second category have the same or similar data quality, and to a certain extent avoiding the problem of unreliable second categories when the target data is affected by noise interference.

[0045] Furthermore, the generalization interval of each patient's target data and the second category of the target data are re-derived based on the second category of the reference data. Within the second category to which each patient's target data belongs, the offset error of each patient's target data is obtained. Based on the offset error of the target data for all patients, abnormal data within the second category are obtained. This process, by re-deriving the generalization interval of each patient's target data and the second category of the target data, updates the generalization interval and second category, further ensuring the accuracy of abnormal data acquisition.

[0046] Finally, for the target data of the newly added patients, the quality of the target data of the newly added patients is judged based on the abnormal data in the second category and whether the target data of the newly added patients falls into the generalization interval of the abnormal data; ensuring that the data quality of the target data of the newly added patients can be obtained in a timely and accurate manner.

[0047] In summary, the present invention uses the generalized interval and second category of the recorded nursing data to timely and accurately judge the quality of the target data of newly added patients, while further avoiding the influence of noise caused by individual differences among patients, screening out low-quality nursing data, and improving the quality of nursing data as a whole. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0049] Figure 1 A framework diagram of an electronic information quality control system for intelligent clinical nursing provided by one embodiment of the present invention;

[0050] Figure 2 This is a flowchart of the steps of the data analysis module in the electronic information quality control system for intelligent clinical nursing provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the electronic information quality control system for intelligent clinical nursing proposed in accordance with the present invention, its specific implementation, structure, features and effects are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0052] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0053] The specific scheme of the electronic information quality control system for intelligent clinical nursing provided by the present invention is described in detail below with reference to the accompanying drawings.

[0054] See also Figure 1 , which shows a framework diagram of an electronic information quality control system for intelligent clinical nursing provided by one embodiment of the present invention. The system includes the following modules:

[0055] The data collection module is used to record relevant data of clinical patients, including patient identity information such as name, gender, age, weight, etc., as well as nursing data such as medical staff information and attending physician information, illness details, medication data, nursing monitoring results, etc.

[0056] Since disease details, medication data, and nursing monitoring results have the greatest impact on the quality of electronic information of clinical nursing, this embodiment only analyzes these three types of nursing data: disease details, medication data, and nursing monitoring results.

[0057] Medical details refer to the specific medical condition summarized by the doctor after consultation and diagnosis using medical equipment (such as CT scans). Medication data refers to the medications used by the patient during clinical care, including drug name, dosage, and duration of use. Nursing monitoring results refer to the test results and time of the patient's physical signs or physiological status during clinical care, such as whether the heart rate, blood pressure, body temperature, and respiratory rate are normal, as well as whether blood and urine test results are normal. All nursing data for each patient is recorded in a MySQL database.

[0058] The data quality of all nursing data for each patient (also known as the electronic information quality mentioned above) mainly refers to: whether the patient's nursing data records are complete; whether multi-source data is properly integrated, such as the diagnosis results from the attending physician, medication data from nurses, and diagnostic monitoring data from medical devices; whether there are conflicts or errors in patient information recorded by different medical personnel; whether the clinical care or treatment process is reasonable, such as whether there is excessive medication or drug substitution for patients, and whether there are inappropriate treatment methods or nursing methods. All of these will affect the data quality of nursing data.

[0059] Data analysis module, used to process nursing data.

[0060] It should be noted that for different patients, the doctors or nurses who register or enter the nursing data are different, and the nursing data may not have a unified format. In order to facilitate the analysis and processing of nursing data, this embodiment uses text-formatted data to describe the disease details when registering or entering nursing data. Professional terms are used to describe the type of disease, and words such as "mild", "severe", and "normal" are used to describe the severity of the disease. Since using text-formatted data to describe the disease is a conventional and commonly used technical means, this embodiment will not go into details. This embodiment uses one-hot encoding technology to encode the disease details into vectors. One-hot encoding technology is a well-known technology and will not be described in detail in this embodiment. Medication data and nursing monitoring results are uniformly recorded using data wide tables; wherein data wide tables are data formats that may be required to be input in the fields of artificial intelligence and machine learning, and are well-known technologies, and will not be described in detail in this embodiment.

[0061] In other embodiments, considering that nursing data is important electronic information in clinical nursing, one of its important functions is to serve as a data set for training machine learning models or neural networks. Therefore, the nursing data can be converted into vectors for application in machine learning models or neural networks. For example, the disease details, medication data, and nursing monitoring results are converted into vectors respectively. Methods for converting nursing data into vectors include word2Vec technology, LSTM neural network, etc. These technologies are well known and are not specifically limited in this embodiment.

[0062] In still other embodiments, templated forms may be used to record and store nursing data.

[0063] The quality control module is used to monitor the quality of nursing data entered into the database. If the nursing data is missing, wrong, or contradictory, it means that there are quality problems in the nursing data (that is, low-quality nursing data).

[0064] Specifically, such as Figure 2As shown, the steps included in the data analysis module are as follows:

[0065] Step S1: any kind of nursing data is recorded as target data; nursing data other than the target data is recorded as reference data and sample data in sequence; the reference data of all patients are clustered to obtain the first category of reference data.

[0066] Among them, the nursing data other than the target data are recorded as reference data and sample data in sequence, specifically including: when the target data is disease details, the reference data and sample data are medication data and nursing monitoring results, respectively; when the target data is medication data, the reference data and sample data are disease details and nursing monitoring results, respectively; when the target data is nursing monitoring results, the reference data and sample data are disease details and medication data, respectively.

[0067] First, the similarity between any two patients' reference data is obtained, denoted as x, and exp(-x) is denoted as the metric distance between the two patients' reference data. The K-Means clustering algorithm is then used to cluster all patients' reference data based on the metric distance to obtain K first categories. Clustering using the K-Means clustering algorithm given a metric distance is a well-known technique and is not described in detail in this embodiment.

[0068] All patients mentioned herein refer to all patients whose clinical care has been completed and whose care data has been entered into the database.

[0069] Considering that the nursing data of different departments (such as respiratory department and orthopedic department) are quite different, all patients in these departments cannot be analyzed together or the same number of data cannot be used to record nursing data. Therefore, in some embodiments, all the patients mentioned above are limited to all patients in each department.

[0070] As an example, the similarity of reference data of any two patients is obtained by:

[0071] When the reference data are disease details, each reference data is represented as a vector (obtained by one-hot encoding technology), and the cosine similarity of the reference data of any two patients is taken as the similarity of the reference data of the two patients.

[0072] When the reference data is medication data or nursing monitoring results, each reference data is represented as a data wide table. For the data wide tables corresponding to the reference data of any two patients, the difference between the table contents at the same position in the two data wide tables is calculated, and the sum of the absolute values ​​of the differences between all table contents is Dx. 1 / (Dx+1) is taken as the similarity of the reference data of the two patients. The larger Dx is, the greater the difference in the contents of the two data wide tables, and the smaller the similarity.

[0073] It should be noted that some tables in a data wide table may record text instead of numerical values. When the text recorded in the tables at the same position in two data wide tables is the same, the difference is recorded as 0; otherwise, the difference is recorded as 1.

[0074] As another example, obtaining the similarity of reference data of any two patients includes the following method:

[0075] When the reference data is represented as a vector (for example, converted into a vector using word2Vec technology), the cosine similarity of the reference data of any two patients is used as the similarity.

[0076] As an example, a method for setting K is: setting K to a preset parameter, for example, K is equal to one twentieth of the number of all patients.

[0077] As another example, K is set by first clustering all patient reference data using the mean-shift clustering algorithm, and using the resulting number of clusters as the value of K. The mean-shift clustering algorithm is a well-known technique that automatically determines the number of clusters, which is then used as the output of the K-Means clustering algorithm (i.e., the K value).

[0078] It should be noted that when the target data is disease details, the reference data and sample data are medication data and nursing monitoring results respectively; when the target data is medication data, the reference data and sample data are disease details and nursing monitoring results respectively; when the target data is nursing monitoring results, the reference data and sample data are disease details and medication data respectively.

[0079] Step S2: Obtain the generalization interval of the target data and the second category of the target data of each patient according to the first category of the reference data.

[0080] S201 : Obtain a generalization interval of target data for each patient according to the first category of reference data.

[0081] The first category contains several reference data with similar characteristics. For example, if the reference data is medication data, the medication data in each first category describes the same or similar medication situations, while the medication data in different first categories have significant differences and distinctions. For another example, if the reference data is medical details, the medical details in each first category represent the same or similar medical conditions, while the medical conditions in different first categories have significant differences and distinctions.

[0082] Each reference data in each first category of reference data corresponds to a patient, and the patient also corresponds to target data and sample data. Therefore, each first category includes both reference data of several patients and target data and sample data of several patients.

[0083] For all target data in any first category of the reference data, any one of which is denoted as A. For each target data in the first category other than target data A, the similarity between target data A and each other target data is obtained. The N target data with the greatest similarity are regarded as the most similar data to target data A. The most similar data to target data A describe multiple other target data that are relatively similar to A.

[0084] In this embodiment, the value of N is half of the number of target data included in each first category (needs to be rounded up); in other embodiments, N can be set to other values, which is not specifically limited in this embodiment.

[0085] The method for obtaining the similarity between the target data A and each other target data is the same as the method for obtaining the similarity between the reference data of any two patients in step S1 , and will not be described in detail in this embodiment.

[0086] Each most similar data corresponds to a sample data. For all sample data corresponding to all the most similar data, the differences between these sample data are obtained. The larger the difference, the greater the difference between these sample data, indicating that multiple other target data with similar differences from A have different sample data, which further indicates that target data A has strong generalization ability. The smaller the difference, the smaller the difference between these sample data, indicating that multiple other target data with similar differences from A have similar sample data, which further indicates that target data A has weaker generalization ability.

[0087] For example, when the target data is disease details, the reference data and sample data are medication data and nursing monitoring results, respectively. The greater the difference between these sample data (that is, the greater the difference), it means that for patients with similar medication data (that is, patients in each first category), N patients with similar or identical disease conditions have relatively different nursing monitoring results. Therefore, it means that the target data A (that is, disease details) has a strong generalization ability. The strong generalization ability means that under the disease represented by the disease details (that is, target data A), when the same or similar drugs are used for treatment, diverse physiological phenomena are exhibited (that is, diverse nursing monitoring results), which further indicates that the target data A is greatly affected by noise interference (for example, it is greatly affected by individual differences in patient weight, age, physiological condition, etc.). If the difference between these sample data is smaller (that is, the smaller the difference), it means that for patients with similar medication data (that is, patients in each first category), N patients with similar or identical illnesses have similar nursing monitoring results. Therefore, it means that the generalization ability of the target data A (that is, the illness details) is weak. The weak generalization ability means that under the disease represented by the illness details (that is, the target data A), when the same or similar drugs are used for treatment, the same or similar physiological phenomena are exhibited (that is, the nursing monitoring results are clear and single), which further indicates that the target data A is less affected by noise interference (for example, it is less affected by individual differences in patient weight, age, physiological condition, etc.).

[0088] As an example, for all sample data corresponding to all the most similar data, obtaining the differences between these sample data includes the following steps:

[0089] The similarity between any two samples of the most similar data is calculated. The mean of the similarities between all samples is denoted by b, and exp(-b) is the difference between these samples. The larger the b, the more similar or identical the samples are, and the smaller the difference between them.

[0090] The method for obtaining the similarity between the sample data is the same as the method for obtaining the similarity between the reference data of any two patients in step S1 , and will not be described in detail in this embodiment.

[0091] Furthermore, based on the aforementioned differences, a generalization interval for each patient's target data A is obtained; this generalization interval represents the range of generalization capabilities exhibited by the target data A under the same or similar reference data (i.e., within the same first category). The larger the range of demonstrated generalization capabilities (i.e., the wider the generalization interval), the higher the data quality of target data corresponding to other reference data in the same first category (hereinafter referred to as "other target data") can be judged to be similar to that of target data A, even if the other target data differ significantly from target data A. If the range of demonstrated generalization capabilities is smaller (i.e., the narrower the generalization interval), the higher the data quality of the other target data can be judged to be significantly or slightly different from that of target data A, when the other target data differ significantly or slightly from target data A.

[0092] As an example, the generalization interval of the target data A of each patient is obtained according to the above-mentioned differences, including the following method:

[0093] For all sample data corresponding to the most similar data of target data A, the difference between these sample data is recorded as a;

[0094] The mean similarity between target data A and all the most similar data is recorded as wa; in the first category to which target data A belongs, all target data other than target data A are obtained, and the mean similarity between target data A and all other target data is recorded as the coordinate value of target data A, recorded as h. The coordinate value h of target data A is used to describe: the position of target data A relative to other target data in the same first category.

[0095] The generalization interval of target data A is recorded as [(1-a / wa)×h, (1+a / wa)×h]; the larger a / wa is, the larger the difference between all sample data corresponding to the most similar data is, indicating that the sample data corresponding to other target data similar to target data A have obvious differences from the sample data corresponding to target data A, and the larger the range of the generalization interval of target data A is at this time; the smaller a / wa is, the smaller the difference between all sample data corresponding to the most similar data is, indicating that the sample data corresponding to other target data similar to target data A have no obvious differences from the sample data corresponding to target data A, and the generalization ability of target data A is relatively weak, and the range of the generalization interval of target data A is smaller.

[0096] As another example, the generalization interval of the target data A of each patient is obtained based on the above-mentioned differences, including the following method:

[0097] In the first category to which the target data A belongs, any target data in the first category is recorded as target data to be processed;

[0098] Obtain similarities between the target data to be processed and other target data in the first category, wherein these similarities constitute a vector, which is the similarity vector of the target data to be processed;

[0099] For all target data in the first category to which target data A belongs, the PCA dimensionality reduction algorithm is used to reduce the similarity vectors of all target data to one dimension, where the dimensionality reduction result of the similarity vector of target data A is taken as h, and then the generalization interval of target data A is recorded as [(1-a / wa)×h, (1+a / wa)×h].

[0100] S202 : Clustering the target data of all patients using the similarity between the target data and the generalization interval to obtain a plurality of second categories of the target data.

[0101] The target data in the same second category not only have the same or similar characteristics, but also any target data is as close as possible to the generalization range of another target data, ensuring that all target data in each second category have the same or similar data quality, and to a certain extent avoiding the problem of unreliable second categories when the target data is affected by noise interference.

[0102] As an example, the target data of all patients are clustered using the similarity between the target data and the generalization interval to obtain several second categories of the target data, including the following specific methods:

[0103] The similarity of the target data of any two patients is recorded as y1. The intersection-and-union ratio of the generalization intervals of the target data of any two patients is recorded as y2, where the intersection-and-union ratio refers to the ratio of the intersection width to the union width of the two generalization intervals.

[0104] Let Y = (1 + y2 × y3) × y1 be the corrected similarity of the target data of the two patients.

[0105] Among them, for the target data Q1 and Q2 of any two patients, when the coordinate value of Q1 falls within the generalization interval of Q2, and the coordinate value of Q2 falls within the generalization interval of Q1, the parameter y3=1, otherwise the parameter y3=0.

[0106] The larger the intersection-over-union ratio of the generalization intervals of the target data for any two patients, and the greater the corrected similarity when either target data falls within the generalization interval of the other, the smaller it is otherwise. Compared to similarity, corrected similarity more accurately describes the similarity in data quality between target data, to a certain extent avoiding noise interference caused by individual patient differences (such as differences in age, weight, and physiological conditions).

[0107] exp(-Y) is used as the metric distance between the target data of any two patients. This metric distance is then used to perform K-Means clustering on the target data of all patients to obtain several second categories of the target data. The above method of this embodiment ensures that any target data in each second category is as close to the generalization interval of another target data as possible, thereby increasing the accuracy of the second category.

[0108] Step S3: For all nursing data of each patient, obtain the generalization interval and the second category of each nursing data.

[0109] When the target data is disease details, the reference data and sample data are medication data and nursing monitoring results, respectively; at this time, according to the methods of steps S1 and S2, the first category of medication data is first obtained, and then the generalization interval of each patient's disease details and several second categories of disease details are obtained based on the first category of medication data.

[0110] When the target data is medication data, the reference data and sample data are disease details and nursing monitoring results, respectively. At this time, according to the method of steps S1 and S2, the first category of disease details is first obtained, and then based on the first category of disease details, the generalization interval of each patient's medication data and several second categories of medication data are obtained.

[0111] When the target data is the nursing monitoring result, the reference data and the sample data are the disease details and the medication data, respectively; at this time, according to the method of steps S1 and S2, the first category of the disease details is first obtained, and then based on the first category of the disease details, the generalization interval of the nursing monitoring results of each patient and several second categories of the nursing monitoring results are obtained.

[0112] So far, for all nursing data of each patient, the generalization interval of each nursing data of each patient and all second categories of each nursing data are obtained; all second categories of each nursing data also include the second category of the reference data.

[0113] Step S4: obtaining the generalization interval and the second category of the target data of each patient again according to the second category of the reference data; obtaining the offset error of the generalization interval of each target data in the second category to which the target data of each patient belongs.

[0114] First, it should be noted that the purpose of the generalization interval for each type of nursing data obtained in step S3 is: when one type of nursing data is similar or close (i.e., within the first category of the reference data in step S2), the generalization interval determines the relationship between the differences between the other nursing data (i.e., between the target data A described in step S2 and the other target data) and the differences in data quality. In other words, when the generalization interval is large, even if the other nursing data have large differences, the data quality of the other nursing data can be judged to be similar. When the generalization interval is small, even if the other nursing data have large or small differences, the data quality of the other nursing data can be judged to be large or small. This avoids interference caused by noise when judging the differences in data quality of nursing data.

[0115] However, the generalization interval of each type of nursing data in step S3 is obtained based on the first category of the reference data in step S1. When the reference data is clustered in step S1, the reference data is interfered by noise, resulting in unreliable first category of the reference data. The unreliable first category means that although some reference data have obvious differences, there are also some differences in the nursing data of the corresponding patients, but such differences are only caused by noise interference. For example, differences in patient weight or physiological conditions lead to differences in medication dosage, blood or urine test results or disease descriptions, but the data quality of the nursing data corresponding to the nursing methods or treatment methods required by these patients is the same. This results in the unreliable classification result of the first category obtained in step S1, and it is impossible to further accurately obtain the most similar data of the target data A, and it is impossible to accurately obtain the generalization interval of the target data A.

[0116] In step S3, the second category of each nursing data has been obtained, which also includes the second category of the reference data described in step S2; that is, the second category of the reference data obtained in step S3, specifically:

[0117] When the target data is disease details, the reference data and sample data are medication data and nursing monitoring results, respectively; at this time, according to the methods of steps S1 and S2, the first category of medication data is first obtained, and then the generalization interval of each patient's disease details and several second categories of disease details are obtained based on the first category of medication data.

[0118] When the target data is medication data, the reference data and sample data are disease details and nursing monitoring results, respectively. At this time, according to the method of steps S1 and S2, the first category of disease details is first obtained, and then based on the first category of disease details, the generalization interval of each patient's medication data and several second categories of medication data are obtained.

[0119] When the target data is the nursing monitoring result, the reference data and the sample data are the disease details and the medication data, respectively; at this time, according to the method of steps S1 and S2, the first category of the disease details is first obtained, and then based on the first category of the disease details, the generalization interval of the nursing monitoring results of each patient and several second categories of the nursing monitoring results are obtained.

[0120] In summary, the following results were obtained for each type of nursing data: generalized intervals and several secondary categories for illness details, generalized intervals and several secondary categories for medication data, and generalized intervals and several secondary categories for nursing monitoring results. Thus, the second categories for each type of nursing data were obtained.

[0121] Now, any type of nursing data is recorded as target data again; nursing data other than target data is recorded as reference data and sample data in turn. For reference data, whether the reference data is disease details, medication data, or nursing monitoring results, the second category corresponding to the reference data can be obtained.

[0122] Therefore, the second category of each type of nursing data includes the second category of reference data.

[0123] In this embodiment, the first category of the reference data described in step S2 is replaced by the second category of the reference data obtained in step S3; the generalization interval of the target data of each patient and the second category of the target data are again obtained using the method in step S2, thereby achieving the update of the generalization interval of the target data of each patient and the update of the second category of the target data.

[0124] When the target data is disease details, the reference data and sample data are medication data and nursing monitoring results, respectively; the above achieves the update of the generalized interval of each patient's disease details and the second category of disease details. When the target data is medication data, the reference data and sample data are disease details and nursing monitoring results, respectively; the above achieves the update of the generalized interval of each patient's medication data and the second category of medication data. When the target data is nursing monitoring results, the reference data and sample data are disease details and medication data, respectively; the above achieves the update of the generalized interval of each patient's nursing monitoring results and the second category of nursing monitoring results.

[0125] In summary, the above process obtains the generalization interval of each patient's target data and the second category of the target data again according to the second category of the reference data, thereby updating the generalization interval of each nursing data of each patient and the second category of each nursing data.

[0126] Next, the first category of the reference data described in step S2 is replaced with the updated second category of the reference data. The method in step S2 is used to again obtain the generalization interval of the target data and the second category of the target data for each patient. This process again updates the generalization interval of the target data and the second category of the target data for each patient. This process is repeated repeatedly, and the generalization interval of the target data and the second category of the target data for each patient are again obtained based on the second category of the reference data (i.e., the updated second category).

[0127] After each execution of this process, that is, after the generalization interval of each patient's target data and the second category of the target data are updated, the generalization intervals of all target data in the second category to which the target data of each patient belong are obtained, and the offset error of the generalization interval of each target data is obtained among the generalization intervals of these target data. The offset error describes the difference between the generalization interval of each target data and the generalization intervals of other target data in the same second category.

[0128] As an example, obtaining the generalization intervals of all target data and obtaining the offset error of the generalization interval of each target data among the generalization intervals of the target data includes the following method:

[0129] For the target data in the second category, any generalization interval among all the generalization intervals of the target data is recorded as B, and all generalization intervals other than B are recorded as reference intervals. The difference B1 between the left endpoint of B and the left endpoint of each reference interval is obtained, and the difference B2 between the right endpoint of B and the endpoint of each reference interval is obtained. [B1, B2] is recorded as the error feature between each reference interval and B. The error features between all reference intervals and B are reduced to one dimension using PCA, and the mean of all dimensionality reduction results is recorded as the offset error of the generalization interval B.

[0130] The set of offset errors of the generalization intervals of all target data in the second category is recorded as the error distribution after each execution of this process.

[0131] Obtain the error distributions of the last three executions of the above process. When the variances and means of the three error distributions are the same, stop executing the above process. At the same time, stop executing the above process after executing the above process multiple times (for example, after 6 times).

[0132] S5. Obtain abnormal data according to the offset error of the generalization interval of each target data in the second category.

[0133] After execution is stopped, the set of offset errors of the generalization intervals of all target data in the second category is recorded as the final error distribution. The final error distribution includes the offset error of the generalization interval of each target data. The final error distribution represents the distribution of noise (that is, the distribution of individual differences in patients such as age, weight, and physiological conditions); however, considering that each type of nursing data of each patient used in steps S1 to S4 refers to the nursing data that has been entered into the data, there may be low-quality nursing data in these data. At this time, the above-mentioned noise distribution also includes the noise caused by low-quality nursing data.

[0134] This embodiment takes into account that among the nursing data of a large number of patients in the database, low-quality nursing data accounts for a minority, so the noise caused by the nursing data is abnormal in the final error distribution. This embodiment uses the LOF anomaly detection algorithm to obtain the abnormal offset error in the final error distribution, and the target data corresponding to the offset error is recorded as abnormal data.

[0135] In the quality control module, abnormal data is regarded as detected low-quality data. When clinical care for certain patients ends, all nursing data of these patients are recorded as nursing data of newly added patients. The quality of the nursing data of the newly added patients is judged based on the detected low-quality data and the generalization interval and second category of the target data after stopping execution in step S4.

[0136] As an example, the method for judging the quality of nursing data for newly added patients includes:

[0137] For all second categories of target data after stopping execution in step S4; record the target data in the nursing data of the newly added patient as data F, calculate the similarity between data F and each target data in the second category, and record the average of the similarities between data F and all target data in the second category as the membership of the second category; take the category with the largest membership as the second category to which data F belongs.

[0138] For any abnormal data in the second category to which data F belongs, the similarity between data F and the abnormal data is recorded as the abnormal value of data F and the abnormal data. When the abnormal value falls into the generalization interval of the abnormal data, the abnormal data is recorded as the target abnormal data, otherwise it is recorded as invalid abnormal data.

[0139] The target data other than all abnormal data in the second category to which data F belongs is recorded as normal data, and the similarity between data F and normal data is recorded as the normal value of data F and normal data. When the normal value falls into the generalization interval of normal data, the normal data is recorded as target normal data, otherwise it is recorded as invalid normal data.

[0140] Obtain the intersection-and-union ratio of the generalization intervals of any pair of target abnormal data and target normal data, and record it as a normal indicator of data F. For all target abnormal data and target normal data, obtain all normal indicators of data F. When the maximum value of the normal indicator is less than the first preset threshold th1, the target data in the nursing data of the newly added patient is recorded as low-quality data, otherwise it is not regarded as low-quality data.

[0141] Specifically, when there is no target anomaly data (i.e., the outlier value does not fall within the generalization interval of any anomaly data) but there is target normal data, data F is no longer considered low-quality data. When there is no target normal data (i.e., the normal value does not fall within the generalization interval of any normal data) but there is target anomaly data, data F is directly considered low-quality data. If neither target anomaly data nor low-quality data exists, data F is also directly considered low-quality data.

[0142] When the target data is medical details, the above method is used to determine whether the medical details in the newly added patient's nursing data are low-quality data. When the target data is medication data, the above method is used to determine whether the medication data in the newly added patient's nursing data are low-quality data. When the target data is nursing monitoring results, the above method is used to determine whether the nursing monitoring results in the newly added nursing data are low-quality data.

[0143] In this embodiment, as long as there is one low-quality data among all the nursing data of each patient, all the nursing data of the patient are abnormal.

[0144] So far, in this embodiment, whenever nursing data of a new patient is added, it is possible to obtain whether the nursing data of the new patient is of low quality.

[0145] When low-quality nursing data is present, all nursing data for that patient is labeled "low quality." In this implementation, nursing data labeled "low quality" is not used as a dataset for training machine learning or neural network models, which are used to implement intelligent clinical care (e.g., developing appropriate care plans).

[0146] It should be noted that this embodiment is described using th1=0.4 as an example. In other embodiments, it can be set to other values. For example, in some embodiments, in order to ensure that the data set for training the machine learning model or the neural network model has high data quality, the value of th1 is set to a larger value, such as th1=0.55; in other embodiments, if the data set for training the machine learning model or the neural network model includes large-scale nursing data, the value of th1 is set to a smaller value, such as th1=0.3. For another example, in some embodiments, when the number of all patients is less than 1000, th1=0.3 is set; when the number of all patients is greater than or equal to 1000 and less than 6000, th1=0.4 is set; when the number of all patients is greater than or equal to 6000, th1=0.55 is set.

[0147] In other embodiments, when low-quality data appears, a manual check is performed, and professional medical staff re-determine whether the low-quality data has obvious missing, incomplete, unreasonable, etc. If so, it will be marked with a "low quality" label. If not, all nursing data of the patient will no longer be regarded as low-quality data (or abnormal data) in all embodiments of the present invention.

[0148] In some other embodiments, after the nursing data of a new patient is entered into the database, all target normal data of the nursing data of the new patient are displayed for reference. Medical staff can judge whether there are any abnormalities in the nursing data of the new patient (such as missing, incomplete, unreasonable, etc.) by comparing the nursing data of the new patient with the target normal data.

[0149] It should be noted that as the patient's nursing data is entered (that is, entered into the database), the nursing data in the database becomes more and more, and the quality judgment results of the nursing data become more and more credible; for the low-quality nursing data (recorded as nursing data Q) obtained after a certain entry of nursing data, after subsequent entry of nursing data, part of the nursing data Q may no longer be of low quality. The reason is that when the nursing data Q is obtained, there is less nursing data in the database, resulting in unreliable nursing data Q. At this time, the low-quality nursing data obtained after the subsequent entry of nursing data is used as the final reliable low-quality nursing data. For the nursing data in the nursing data Q that is no longer of low quality, its "low quality" label is deleted.

[0150] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An electronic information quality control system for intelligent clinical nursing, characterized in that: The system includes the following modules: The data collection module is used to obtain several types of nursing data for each patient, including: disease details, medication data, and nursing monitoring results; The data analysis module is used to record any kind of nursing data as target data; the nursing data other than the target data are recorded as reference data and sample data in sequence. When the target data is disease details, the reference data and sample data are medication data and nursing monitoring results, respectively; when the target data is medication data, the reference data and sample data are disease details and nursing monitoring results, respectively; when the target data is nursing monitoring results, the reference data and sample data are disease details and medication data, respectively; the reference data of all patients are clustered to obtain the first category of the reference data; Obtaining the generalization interval of target data of each patient and the second category of target data according to the first category of reference data, including: obtaining a number of most similar data of the target data of each patient in the first category of reference data, and obtaining the generalization interval of the target data of each patient according to the difference between the sample data corresponding to the number of most similar data; clustering the target data of all patients by using the similarity between the target data, the generalization interval and whether the target data is within the generalization interval, to obtain a number of second categories of target data; For all nursing data of each patient, a generalization interval of all nursing data of each patient and a second category of each nursing data are obtained; the second category of each nursing data includes a second category of reference data; Update the generalization interval of the target data of each patient and the second category to which the target data belongs according to the second category of the reference data; obtain the offset error of the target data of each patient in the second category to which the target data of each patient belongs; obtain the abnormal data in the second category to which the target data belongs according to the offset errors of the target data of all patients; The quality control module determines the quality of the target data of the newly added patient according to the abnormal data in the second category to which the target data belongs and whether the target data of the newly added patient falls into the generalization interval of the abnormal data.

2. The electronic information quality control system for intelligent clinical nursing according to claim 1, characterized in that: The clustering of the reference data of all patients to obtain the first category of reference data includes the following specific steps: The similarity of the reference data of any two patients is obtained, denoted as x, and exp(-x) is denoted as the metric distance of the reference data of any two patients. The reference data of all patients are clustered using the K-Means clustering algorithm according to the metric distance to obtain several first categories, and exp() represents an exponential function with a natural constant as the base.

3. The electronic information quality control system for intelligent clinical nursing according to claim 1 is characterized in that: The step of obtaining a plurality of most similar data of the target data of each patient and obtaining a generalization interval of the target data of each patient according to the difference between the sample data corresponding to the plurality of most similar data includes the following specific steps: Each first category of the reference data includes target data corresponding to the reference data of each patient; For all target data in any first category of the reference data, the target data of any patient is recorded as A. For each target data other than the target data A in the first category, the similarity between the target data A and each other target data is obtained, and the target data with the greatest similarity are taken as the most similar data of the target data A; For all sample data corresponding to the most similar data of target data A, the difference between all sample data is recorded as a; The mean similarity between the target data A and all the most similar data is recorded as wa; In the first category to which the target data A belongs, the average value of the similarity between the target data A and all other target data is obtained and recorded as the coordinate value of the target data A, which is recorded as h; The generalization interval of the target data A is recorded as [(1-a / wa)×h, (1+a / wa)×h].

4. The electronic information quality control system for intelligent clinical nursing according to claim 3 is characterized in that: The specific steps for obtaining the differences between all sample data are as follows: In all the sample data, the similarity between any two sample data is obtained, the mean of the similarities between all the sample data is recorded as b, exp(-b) is recorded as the difference between all the sample data, and exp() represents an exponential function with a natural constant as the base.

5. The electronic information quality control system for intelligent clinical nursing according to claim 3 is characterized in that: The method of clustering the target data of all patients by using the similarity between the target data, the generalization interval, and whether the target data is within the generalization interval to obtain several second categories of the target data includes the following specific steps: The similarity of the target data of any two patients is recorded as y1; the intersection and union ratio of the generalization intervals of the target data of any two patients is recorded as y2; Let Y = (1 + y2 × y3) × y1 be the corrected similarity of the target data of any two patients; Among them, for any two patients' target data Q1 and Q2, when the coordinate value of Q1 falls within the generalization interval of Q2, and the coordinate value of Q2 falls within the generalization interval of Q1, the parameter y3=1, otherwise the parameter y3=0; exp(-Y) is used as the metric distance between the target data of any two patients, and then the metric distance is used to perform K-Means clustering on the target data of all patients to obtain several second categories of the target data; exp() represents an exponential function with a natural constant as the base.

6. The electronic information quality control system for intelligent clinical nursing according to claim 1, characterized in that: The method includes updating the generalization interval of the target data of each patient and the second category to which the target data belongs according to the second category of the reference data; obtaining the offset error of the target data of each patient in the second category to which the target data of each patient belongs; and obtaining the abnormal data in the second category to which the target data belongs according to the offset error of the target data of all patients, including the following specific steps: H1: Replace the first category of the reference data with the second category of the reference data; then update the generalization interval of the target data of each patient and the second category to which the target data belongs according to the second category of the reference data, and for all the nursing data of each patient, obtain the generalization interval of all the nursing data of each patient and the second category of each nursing data; the second category of each nursing data includes the second category of the reference data; H2: Repeat the steps in H1; After each execution of the steps in H1, the offset error of the generalization interval of each target data in the second category to which the target data of each patient belongs is obtained; the set consisting of the offset errors of the generalization intervals of all target data in the second category to which all target data belong is recorded as the error distribution after each execution of H1; when the error distribution no longer changes, the execution of the steps in H1 is stopped; after stopping the execution of the steps in H1, the abnormal data in the second category to which the target data belongs is obtained according to the error distribution.

7. The electronic information quality control system for intelligent clinical nursing according to claim 6, characterized in that: The specific steps of obtaining the offset error of the generalization interval of each target data are as follows: For the target data in the second category to which the target data belongs, any generalization interval among the generalization intervals of all the target data is recorded as B, and the other generalization intervals other than B are recorded as reference intervals. The difference B1 between the left endpoint of B and the left endpoint of each reference interval is obtained, and the difference B2 between the right endpoint of B and the right endpoint of each reference interval is obtained. [B1, B2] is recorded as the error feature between each reference interval and B. The error characteristics of all reference intervals and B are subjected to PCA dimensionality reduction, and the mean of all dimensionality reduction results is recorded as the offset error of the generalized interval B.

8. The electronic information quality control system for intelligent clinical nursing according to claim 6, characterized in that: The specific steps of obtaining abnormal data in the second category to which the target data belongs according to the error distribution include the following: The final error distribution includes the offset error of the generalization interval of each target data; the LOF anomaly detection algorithm is used to obtain the abnormal offset error in the final error distribution, and the target data corresponding to the abnormal offset error is recorded as abnormal data.

9. The electronic information quality control system for intelligent clinical nursing according to claim 1, characterized in that: The specific steps of judging the quality of the target data of the newly added patient according to the abnormal data in the second category to which the target data belongs and whether the target data of the newly added patient falls into the generalization interval of the abnormal data are as follows: Record the target data of the newly added patient as data F, and obtain the second category to which data F belongs from among all the second categories of the target data; For any abnormal data in the second category to which data F belongs, the similarity between data F and the abnormal data is recorded as the abnormal value of data F and the abnormal data. When the abnormal value falls into the generalization interval of the abnormal data, the abnormal data is recorded as the target abnormal data; The target data other than all abnormal data in the second category to which data F belongs is recorded as normal data, and the similarity between data F and normal data is recorded as the normal value of data F and normal data. When the normal value falls into the generalization interval of normal data, the normal data is recorded as the target normal data; The quality of the target data of the newly added patients is judged based on the generalization interval between the target abnormal data and the target normal data.

10. The electronic information quality control system for intelligent clinical nursing according to claim 9, characterized in that: The method of judging the quality of the target data of the newly added patient according to the generalization interval of the target abnormal data and the target normal data includes the following specific steps: Obtain the intersection and union ratio of the generalization intervals of any pair of target abnormal data and target normal data, and record it as a normal indicator of data F. For all target abnormal data and target normal data, obtain all normal indicators of data F. When the maximum value of the normal indicator is less than the first preset threshold th1, the target data of the newly added patient is recorded as low-quality data, otherwise it is not recorded as low-quality data.

Images