Serum potassium real-time quality control model establishment and application method based on pre-classification and related device

Abstract

The invention discloses a serum potassium real-time quality control model establishment and application method based on pre-classification and a related device, and relates to the technical field of real-time quality control, and the method comprises the steps: obtaining a first training set, a second training set and a third training set, the first training set, the second training set and the third training set respectively comprise a plurality of historical serum potassium detection data of dialysis patients, healthy physical examination persons and non-dialysis patients, determining a first real-time quality control model based on a plurality of statistical methods and performance comparison by taking the first training set as input, and determining a second real-time quality control model based on a plurality of statistical methods and performance comparison by taking the second training set as input; and determining a second real-time quality control model on the basis of various statistical methods and performance comparison by taking the first training set as input, and determining a third real-time quality control model on the basis of various statistical methods and performance comparison by taking the third training set as input. Therefore, the false positive rate in the serum potassium real-time quality control process can be reduced.

Description

Technical Field

[0001] The present application relates to the technical field of real-time quality control, and in particular to a method for establishing and applying a real-time quality control model for serum potassium based on pre-classification, and related devices. Background Art

[0002] Against the backdrop of continued advancements in clinical diagnostics, patient-based real-time quality control (PBRTQC) has emerged as a key technology dedicated to ensuring the accuracy and reliability of real-time laboratory results. Unlike traditional laboratory quality control methods, which typically rely on periodic control sample testing to monitor system performance, these methods suffer from limitations such as the inability to monitor in real time, limited monitoring during the analytical process, matrix effects of quality control materials, and weak inter-system comparison capabilities. PBRTQC directly utilizes daily patient data, enabling a shift from traditional quality monitoring to a more dynamic, immediate approach. This offers numerous advantages, including the ability to enable continuous real-time monitoring, monitoring of pre-analytical and in-analytical data, lack of matrix effects, and comprehensive comparison capabilities, encompassing both longitudinal comparisons of individual instruments and cross-sectional comparisons of multiple instruments. While PBRTQC offers significant benefits, it also presents the challenge of a high false-positive rate. Summary of the Invention

[0003] The purpose of this application is to provide a method for establishing and applying a real-time quality control model for serum potassium based on pre-classification and related devices, which can reduce the false positive rate in the real-time quality control process of serum potassium.

[0004] To achieve the above objectives, this application provides the following solutions:

[0005] In a first aspect, the present application provides a method for establishing a real-time quality control model for serum potassium based on pre-classification, comprising:

[0006] Obtain a first training set, a second training set, and a third training set; the first training set includes multiple historical serum potassium test data of dialysis patients in a first historical period, the second training set includes multiple historical serum potassium test data of healthy subjects in the first historical period, and the third training set includes multiple historical serum potassium test data of non-dialysis patients in the first historical period;

[0007] Using the first training set as input, modeling is performed using multiple statistical methods to obtain a first model corresponding to each of the statistical methods and the performance of each of the first models;

[0008] Using the second training set as input, modeling is performed using multiple statistical methods to obtain a second model corresponding to each of the statistical methods and the performance of each of the second models;

[0009] Using the third training set as input, modeling is performed using multiple statistical methods to obtain a third model corresponding to each statistical method and the performance of each third model;

[0010] The first model with the best performance is selected as the first real-time quality control model, the second model with the best performance is selected as the second real-time quality control model, and the third model with the best performance is selected as the third real-time quality control model; the first real-time quality control model is used to perform real-time quality control on the serum potassium detection data of dialysis patients, the second real-time quality control model is used to perform real-time quality control on the serum potassium detection data of healthy physical examination subjects, and the third real-time quality control model is used to perform real-time quality control on the serum potassium detection data of non-dialysis patients.

[0011] In a second aspect, the present application provides an application method of a real-time quality control model for serum potassium based on pre-classification, comprising:

[0012] Performing real-time quality control on serum potassium test data of dialysis patients using a first real-time quality control model; the first real-time quality control model is a first real-time quality control model obtained using the above-mentioned method for establishing a real-time quality control model of serum potassium based on pre-classification;

[0013] Performing real-time quality control on the serum potassium test data of healthy subjects using a second real-time quality control model; the second real-time quality control model is obtained using the above-mentioned method for establishing a real-time quality control model of serum potassium based on pre-classification;

[0014] A third real-time quality control model is used to perform real-time quality control on serum potassium test data of non-dialysis patients; the third real-time quality control model is a third real-time quality control model obtained by using the above-mentioned method for establishing a serum potassium real-time quality control model based on pre-classification.

[0015] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned method for establishing a real-time quality control model of serum potassium based on pre-classification or the above-mentioned method for applying a real-time quality control model of serum potassium based on pre-classification.

[0016] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for establishing a real-time quality control model of serum potassium based on pre-classification or the above-mentioned method for applying a real-time quality control model of serum potassium based on pre-classification.

[0017] In a fifth aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements the above-mentioned method for establishing a real-time quality control model of serum potassium based on pre-classification or the above-mentioned method for applying a real-time quality control model of serum potassium based on pre-classification.

[0018] According to the specific embodiments provided in this application, this application has the following technical effects:

[0019] The present application provides a method for establishing and applying a real-time quality control model of serum potassium based on pre-classification and a related device. It is found that the traditional PBRTQC model does not distinguish between population types, and the serum potassium of dialysis patients fluctuates greatly (high before dialysis and low after dialysis), resulting in a high false positive rate of the traditional PBRTQC model. Based on this, a first training set, a second training set and a third training set are obtained. The first training set includes multiple historical serum potassium detection data of dialysis patients in the first historical period, the second training set includes multiple historical serum potassium detection data of healthy people in the first historical period, and the third training set includes multiple historical serum potassium detection data of non-dialysis patients in the first historical period. The first training set is used as input and a variety of statistical methods are used to perform the real-time quality control model. Modeling is performed, and a first real-time quality control model is determined through performance comparison. Subsequently, real-time quality control is performed on the serum potassium detection data of dialysis patients. The second training set is used as input, and modeling is performed using multiple statistical methods. The second real-time quality control model is determined through performance comparison. Subsequently, real-time quality control is performed on the serum potassium detection data of healthy physical examination subjects. The third training set is used as input, and modeling is performed using multiple statistical methods. The third real-time quality control model is determined through performance comparison. Subsequently, real-time quality control is performed on the serum potassium detection data of non-dialysis patients. The present application can establish respective real-time quality control models based on different types of populations, thereby reducing the false positive rate in the serum potassium real-time quality control process. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0021] Figure 1This is an application environment diagram of a method for establishing a real-time quality control model for serum potassium based on pre-classification provided in Example 1 of the present application.

[0022] Figure 2 This is a flow chart of a method for establishing a real-time quality control model for serum potassium based on pre-classification provided in Example 1 of the present application.

[0023] Figure 3 Schematic diagram of the PBRTQC model establishment process provided in Example 1 of this application.

[0024] Figure 4 Schematic diagram of the false positive rate (FPR) of the PBRTQC model for the physical examination group, dialysis group, and non-dialysis group provided in Example 1 of the present application; wherein, Figure 4 (a) is the false positive rate obtained by validating the PBRTQC model of the dialysis group in validation set 1. Figure 4 (b) is the false positive rate obtained by the PBRTQC model of the physical examination group in validation set 1. Figure 4 (c) is the false positive rate obtained by validating the PBRTQC model of the non-dialysis group in validation set 1. Figure 4 (d) is the false positive rate obtained by validating the PBRTQC model of the non-dialysis group in validation set 2. Figure 4 (e) in the figure is the false positive rate obtained by validating the PBRTQC model of the physical examination group in validation set 3.

[0025] Figure 5 Schematic diagram of the validation results of the PBRTQC model for the physical examination group, dialysis group, and non-dialysis group provided in Example 1 of the present application; wherein, Figure 5 (a) in the figure shows the ANPed (timely rate, the average number of affected patient results before error detection) and MNPed (timely rate, the median number of affected patient results before error detection) results obtained by validating the PBRTQC model of the physical examination group, dialysis group, and non-dialysis group using validation set 1, respectively. Figure 5 (b) shows the FPR and APed (average probability of error detection) results obtained by using validation set 1 to validate the PBRTQC model of the physical examination group, dialysis group, and non-dialysis group. Figure 5 (c) in the figure shows the ANPed and MNPed results obtained by using validation set 2 to validate the PBRTQC model of the non-dialysis group, and the ANPed and MNPed results obtained by using validation set 3 to validate the PBRTQC model of the physical examination group. Figure 5 (d) in the figure shows the FPR and Aped results obtained by using validation set 2 to verify the PBRTQC model of the non-dialysis group, and the FPR and Aped results obtained by using validation set 3 to verify the PBRTQC model of the physical examination group.

[0026] Figure 6 This is a flow chart of a method for applying a pre-classification-based serum potassium real-time quality control model provided in Example 2 of the present application.

[0027] Figure 7 A schematic diagram of the structure of a computer device provided in Example 3 of the present application. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0029] Example 1

[0030] The method for establishing a real-time quality control model for serum potassium based on pre-classification provided in the embodiments of the present application can be applied to Figure 1 In the application environment shown, the terminal communicates with the server via a network. The data storage system can store data that the server needs to process. The data storage system can be set up separately, integrated on the server, or placed on a cloud or other server. The terminal can send a pending establishment request to the server. After the server receives the pending establishment request, the server obtains a first training set, a second training set, and a third training set for the pending establishment request; uses the first training set as input and uses multiple statistical methods to perform modeling, obtaining a first model corresponding to each statistical method and the performance of each first model; uses the second training set as input and uses multiple statistical methods to perform modeling, obtaining a second model corresponding to each statistical method and the performance of each second model; uses the third training set as input and uses multiple statistical methods to perform modeling, obtaining a third model corresponding to each statistical method and the performance of each third model; selects the first model with the best performance as the first real-time quality control model, selects the second model with the best performance as the second real-time quality control model, and selects the third model with the best performance as the third real-time quality control model. The server can feedback the establishment results of the first real-time quality control model, the second real-time quality control model, and the third real-time quality control model obtained for the establishment request to the terminal.

[0031] In addition, in some embodiments, the method for establishing a real-time quality control model of serum potassium based on pre-classification can also be implemented separately by a server or a terminal. For example, the terminal can directly process the pending establishment request, or the server can obtain the pending establishment request from the data storage system and process the pending establishment request.

[0032] The terminals may include, but are not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices may include smart speakers, smart TVs, smart air conditioners, and smart car devices. Portable wearable devices may include smart watches, smart bracelets, and head-mounted devices. The server may be implemented as a standalone server or a server cluster consisting of multiple servers, or as a cloud server.

[0033] In an exemplary embodiment, Figure 2 As shown, a method for establishing a real-time quality control model of serum potassium based on pre-classification is provided. The method is executed by a computer device, specifically a computer device such as a terminal or a server, or a terminal and a server. In the embodiment of the present application, the method is applied to Figure 1 The following steps are used as an example to illustrate the server.

[0034] Step S1, obtaining a first training set, a second training set, and a third training set; the first training set includes multiple historical serum potassium test data of dialysis patients in a first historical period, the second training set includes multiple historical serum potassium test data of healthy subjects in the first historical period, and the third training set includes multiple historical serum potassium test data of non-dialysis patients in the first historical period.

[0035] Step S2: using the first training set as input, performing modeling using a plurality of statistical methods, and obtaining a first model corresponding to each of the statistical methods and the performance of each of the first models.

[0036] Step S3: Using the second training set as input, modeling is performed using a plurality of statistical methods to obtain a second model corresponding to each of the statistical methods and the performance of each of the second models.

[0037] Step S4: using the third training set as input, performing modeling using a plurality of statistical methods, and obtaining a third model corresponding to each of the statistical methods and the performance of each of the third models.

[0038] Step S5: Select the first model with the best performance as the first real-time quality control model, select the second model with the best performance as the second real-time quality control model, and select the third model with the best performance as the third real-time quality control model; the first real-time quality control model is used to perform real-time quality control on the serum potassium test data of dialysis patients, the second real-time quality control model is used to perform real-time quality control on the serum potassium test data of healthy subjects, and the third real-time quality control model is used to perform real-time quality control on the serum potassium test data of non-dialysis patients.

[0039] By implementing the above steps S1 to S5, this embodiment found that the traditional PBRTQC model did not distinguish between population types (dialysis patients, healthy physical examination subjects, and non-dialysis patients), and the serum potassium of dialysis patients fluctuated greatly (high before dialysis and low after dialysis), resulting in a high false positive rate. In addition, it was difficult to meet the requirements of the CLSIEP28-A3c guideline for population stratification by modeling based on historical serum potassium test data of mixed population types. Therefore, a method for establishing a real-time quality control model of serum potassium based on pre-classification was provided, and real-time quality control models for different types of populations were established, aiming to reduce the high false positive rate of the PBRTQC model caused by the large fluctuation of serum potassium in dialysis patients and improve the real-time monitoring accuracy of the laboratory detection system.

[0040] like Figure 3 As shown, the method for establishing a real-time quality control model for serum potassium based on pre-classification in this embodiment includes the steps of data collection and classification, data preprocessing, model construction, model verification, implementation and monitoring, etc., which are described in detail as follows.

[0041] Step 1: Data collection and classification.

[0042] In this embodiment, historical serum potassium test data of different types of people are first collected and classified according to specific types of people. Specifically, historical serum potassium test data (i.e., serum potassium concentration) of different types of people are collected from the hospital information system, including historical serum potassium test data of dialysis patients (who are sick and have undergone dialysis treatment), healthy physical examination subjects (who are not sick and have only gone to the hospital for a health checkup), and non-dialysis patients (who are sick but have not undergone dialysis treatment). The historical serum potassium test data of different types of people in a suitable time period (such as within one year) are selected as training data sets for model construction. The training data sets are divided into three groups. The classification is based on whether they are healthy physical examination subjects and whether the patients undergo dialysis treatment. The three groups of training data are: (1) dialysis group training data: historical serum potassium test data of dialysis patients; (2) physical examination group training data: historical serum potassium test data of healthy physical examination subjects; (3) non-dialysis group training data: historical serum potassium test data of non-dialysis patients (all patients except dialysis patients and healthy physical examination subjects). Step 1 ensures the data integrity and pertinence of subsequent analysis.

[0043] At this time, in this embodiment, a first training set (i.e., dialysis group training data), a second training set (i.e., physical examination group training data), and a third training set (i.e., non-dialysis group training data) are obtained. The first training set includes multiple historical serum potassium test data of dialysis patients in the first historical period, the second training set includes multiple historical serum potassium test data of healthy physical examination subjects in the first historical period, and the third training set includes multiple historical serum potassium test data of non-dialysis patients in the first historical period. The length of the first historical period can be determined according to user needs.

[0044] Step 2: data preprocessing.

[0045] Step 2 is to preprocess the three groups of training data obtained by classification in step 1. Specifically, the three groups of training data collected in step 1 are cleaned and outliers are removed. Outliers include historical serum potassium test data of non-serum specimens, historical serum potassium test data that is not within a reasonable range, and historical serum potassium test data that is not within a cutoff range. The reasonable range refers to the medical reasonable range, which can be determined based on user experience. The cutoff ranges corresponding to the three groups of training data are different. For a certain group of training data, the cutoff range of the training data is determined according to the data distribution of the training data. Specifically, a concentration distribution range histogram is drawn for each group of training data. The horizontal axis of the concentration distribution range histogram is the different values ​​of the historical serum potassium test data, and the vertical axis is the number of samples, which expresses the distribution or proportion of samples within a certain concentration range. Based on the concentration distribution range histogram, the optimal cutoff range of the training data is determined based on experience. Through data cleaning, three groups of preprocessed training data are obtained to ensure the accuracy and availability of the three groups of preprocessed training data. At the same time, we further confirmed the classification accuracy of the three sets of preprocessed training data and determined whether the historical serum potassium test data corresponded to the correct population type, providing a clear training dataset for subsequent model construction. Step two ensures the quality of the data input to the model, which is a key prerequisite for improving model performance.

[0046] At this time, in this embodiment, the first training set is used as input, and modeling is performed using multiple statistical methods to obtain the first model corresponding to each statistical method and the performance of each first model; the second training set is used as input, and modeling is performed using multiple statistical methods to obtain the second model corresponding to each statistical method and the performance of each second model; the third training set is used as input, and modeling is performed using multiple statistical methods to obtain the third model corresponding to each statistical method and the performance of each third model.

[0047] The method uses the first training set as input and utilizes multiple statistical methods to perform modeling, thereby obtaining a first model corresponding to each statistical method and the performance of each first model. The method specifically includes: performing data cleaning on the first training set to obtain a first cleaned training set; and using the first cleaned training set as input and utilizing multiple statistical methods to perform modeling, thereby obtaining a first model corresponding to each statistical method and the performance of each first model. The data cleaning includes removing outliers, wherein outliers are historical serum potassium test data obtained from testing non-serum specimens, historical serum potassium test data that is not within a reasonable range, and historical serum potassium test data that is not within a first cutoff range. The first cutoff range is determined based on the data distribution of the historical serum potassium test data in the first training set.

[0048] Similarly, using the second training set as input, modeling is performed using multiple statistical methods to obtain a second model corresponding to each statistical method and the performance of each second model, specifically including: performing data cleaning on the second training set to obtain a second cleaned training set; using the second cleaned training set as input, modeling is performed using multiple statistical methods to obtain a second model corresponding to each statistical method and the performance of each second model. Data cleaning includes removing outliers, which are historical serum potassium test data obtained from testing non-serum specimens, historical serum potassium test data that is not within a reasonable range, and historical serum potassium test data that is not within a second cutoff range. The second cutoff range is determined based on the data distribution of the historical serum potassium test data in the second training set.

[0049] Similarly, using the third training set as input, modeling using multiple statistical methods to obtain a third model corresponding to each statistical method and the performance of each third model specifically includes: performing data cleaning on the third training set to obtain a third cleaned training set; using the third cleaned training set as input, modeling using multiple statistical methods to obtain a third model corresponding to each statistical method and the performance of each third model. Data cleaning includes removing outliers, which are historical serum potassium test data obtained from testing non-serum specimens, historical serum potassium test data that is not within a reasonable range, and historical serum potassium test data that is not within a third cutoff range. The third cutoff range is determined based on the data distribution of the historical serum potassium test data in the third training set.

[0050] The first cleaned training set, the second cleaned training set, and the third cleaned training set are three sets of preprocessed training data.

[0051] Step three: model construction.

[0052] In step three, based on the three sets of preprocessed training data obtained in step two, four statistical methods were used to construct an initial PBRTQC model for each set of preprocessed training data. Based on industry standards, the most appropriate statistical method was selected for each set of preprocessed training data to obtain a corresponding PBRTQC model. By selecting appropriate statistical methods in step three, the resulting PBRTQC model accurately reflects the characteristics of serum potassium testing in different populations.

[0053] For each set of pre-processed training data, in the process of building the PBRTQC model, for each statistical method, we first determine the multiple values ​​of each model parameter of the statistical method, and randomly combine the multiple values ​​of each model parameter to obtain multiple parameter value combinations. Each parameter value combination corresponds to an initial PBRTQC model. The total allowable error (Total Error Allowable (TEA, such as 6%) and the expected false positive rate, each initial PBRTQC model is used to perform real-time quality control on the preprocessed training data to obtain the false positive rate (FPR) corresponding to each initial PBRTQC model, and then an error is added to the preprocessed training data. The error is based on the total allowable error and can be a positive error and / or a negative error. The positive error can be a times the total allowable error, and the negative error can be b times the total allowable error, where a and b are both less than 1. When a positive error is added, each historical serum potassium test data in the preprocessed training data is added with the value corresponding to the positive error (the mean of the preprocessed training data in the group * the positive error). When a negative error is added, each historical serum potassium test data in the preprocessed training data is subtracted from the value corresponding to the negative error (the mean of the preprocessed training data in the group * the negative error). Each initial PBRTQC model is used to perform real-time quality control on the preprocessed training data with added positive and / or negative errors to obtain the false negative rate (False Negative Rate) corresponding to each initial PBRTQC model. Rate, FNR) and timely rate (ANPed). When adding positive errors, the false negative rate is recorded as the positive false negative rate, and the timely rate is recorded as the positive timely rate. When adding negative errors, the false negative rate is recorded as the negative false negative rate, and the timely rate is recorded as the negative timely rate. When adding positive and negative errors, the false negative rate is the sum of the positive false negative rate and the negative false negative rate, and the timely rate is the sum of the positive timely rate and the negative timely rate. Finally, an initial PBRTQC model is selected as the initial PBRTQC model of the statistical method based on the fact that the deviation between the false positive rate and the expected false positive rate is less than the threshold, and the sum of the false negative rate, false positive rate and timely rate is the smallest. Similarly, an initial PBRTQC model of a statistical method is selected as the final PBRTQC model based on the fact that the deviation between the false positive rate and the expected false positive rate is less than the threshold, and the sum of the false negative rate, false positive rate and timely rate is the smallest.

[0054] It should be noted that the PBRTQC model inputs historical serum potassium test data from different samples, and outputs the quality control value calculated by the PBRTQC model. For example, if the PBRTQC model ultimately uses the moving median method with a step size of 35, the quality control value is the median value after sorting the historical serum potassium test data of 35 samples within the cutoff range from smallest to largest. Quality control values ​​that exceed the upper and lower error limits are considered out-of-control quality control values, also known as false-positive quality control values. The false-positive rate = the number of false-positive quality control values ​​ / the total number of quality control values ​​(i.e., the ratio of the number of quality control values ​​exceeding the upper and lower error limits to the total number of quality control values ​​without the introduction of error). The false-negative rate = the number of false-negative quality control values ​​ / the total number of quality control values ​​with the introduction of error (i.e., the ratio of the number of quality control values ​​within the upper and lower error limits after the introduction of error to the total number of quality control values ​​after the introduction of error). Timeliness: Use ANPed, which is the average of the total number of quality control values ​​required to detect the error (a certain quality control value that introduced the error is out of control) after the error is introduced, to evaluate the timeliness of the PBRTQC program in detecting errors.

[0055] At this time, in this embodiment, the first training set is used as input, and multiple statistical methods are used to perform modeling, thereby obtaining a first model corresponding to each statistical method and the performance of each first model; the second training set is used as input, and multiple statistical methods are used to perform modeling, thereby obtaining a second model corresponding to each statistical method and the performance of each second model; the third training set is used as input, and multiple statistical methods are used to perform modeling, thereby obtaining a third model corresponding to each statistical method and the performance of each third model. The first model with the best performance is selected as the first real-time quality control model, the second model with the best performance is selected as the second real-time quality control model, and the third model with the best performance is selected as the third real-time quality control model. The first real-time quality control model is used to perform real-time quality control on serum potassium test data of dialysis patients, the second real-time quality control model is used to perform real-time quality control on serum potassium test data of healthy subjects, and the third real-time quality control model is used to perform real-time quality control on serum potassium test data of non-dialysis patients.

[0056] The first training set is used as input, and multiple statistical methods are used to perform modeling to obtain the first model corresponding to each statistical method and the performance of each first model, specifically including:

[0057] (1) For each statistical method, multiple values ​​of each model parameter corresponding to the statistical method are determined, and the multiple values ​​of each model parameter are randomly combined to generate multiple parameter value combinations, where the parameter value combinations include one value of each model parameter.

[0058] In the present embodiment, the statistical method includes a moving average method, a moving median method, a weighted moving average method, and an exponentially weighted moving average method. When the statistical method is a moving average method, a moving median method, and a weighted moving average method, the model parameters include a step size, an upper error limit, and a lower error limit. When the statistical method is an exponentially weighted moving average method, the model parameters include a step size, an upper error limit, a lower error limit, and a weighting coefficient. Wherein, the step size is the number of serum potassium test data that need to be included when performing a real-time quality control. The upper error limit and the lower error limit are used to determine whether the quality control value obtained by performing a real-time quality control is out of control. When the quality control value is greater than the upper error limit or less than the lower error limit, the quality control value is out of control.

[0059] Specifically, according to the amount of data and the fluctuation characteristics, the appropriate step size, upper error limit and lower error limit are selected. First, the maximum acceptable step size is determined. For example, if the maximum step size has been determined to be 80, then the multiple values ​​of the step size can be 5, 10, 15, 20, ... 75, 80. The upper error limit and the lower error limit are determined according to the mean and standard deviation of the first training set. The multiple values ​​of the upper error limit can be mean + M * standard deviation, M = 0.5, 1, 1.5, 2, ..., 4.5, 5. The multiple values ​​of the lower error limit can be mean - M * standard deviation. The multiple values ​​of the weighting coefficient can be determined according to user needs. Optionally, the model parameters can also include an upper alarm limit and a lower alarm limit. The upper alarm limit and the lower alarm limit are used to determine whether the quality control value obtained by performing a real-time quality control needs to be paid attention to. When the quality control value is greater than the upper alarm limit or less than the lower alarm limit, the quality control value needs to be paid attention to and may be out of control later. The upper alarm limit is less than the upper error limit, and the lower alarm limit is greater than the lower error limit. The upper alarm limit and the lower alarm limit are determined based on the mean and standard deviation of the first training set. The multiple values ​​of the upper alarm limit can be mean + Q * standard deviation, Q = 0.5, 1, 1.5, 2, ..., 4.5, 5. The multiple values ​​of the lower alarm limit can be mean - Q * standard deviation.

[0060] (2) For each parameter value combination, a first initial model is determined based on the parameter value combination, and the first initial model is used to perform real-time quality control on the first training set to obtain the performance of the first initial model.

[0061] The performance includes a false positive rate, a false negative rate, and a timely rate. At this time, the first initial model is used to perform real-time quality control on the first training set to obtain the performance of the first initial model, specifically including: using the first initial model to perform real-time quality control on the first training set to obtain the false positive rate corresponding to the first initial model; based on the total allowable error, adding an error to each historical serum potassium test data in the first training set to obtain the first error training set; using the first initial model to perform real-time quality control on the first error training set to obtain the false negative rate and timely rate corresponding to the first initial model; and using the false positive rate, false negative rate, and timely rate corresponding to the first initial model as the performance of the first initial model.

[0062] (3) Select the first initial model with the best performance as the first model corresponding to the statistical method, and use the performance of the first initial model with the best performance as the performance of the first model to obtain the first model corresponding to each statistical method and the performance of each first model.

[0063] Selecting the first initial model with the best performance as the first model corresponding to the statistical method specifically includes: judging whether the absolute value of the difference between the false positive rate corresponding to the first initial model and the expected false positive rate is less than a preset difference; if so, determining the first initial model as the first post-screening initial model; calculating the sum of the false positive rate, false negative rate and timely rate corresponding to the first post-screening initial model, and selecting the first post-screening initial model with the smallest sum as the first model corresponding to the statistical method.

[0064] Similarly, the second training set is used as input and multiple statistical methods are used to perform modeling. The second model corresponding to each statistical method and the performance of each second model are obtained, specifically including:

[0065] (1) For each statistical method, multiple values ​​of each model parameter corresponding to the statistical method are determined, and the multiple values ​​of each model parameter are randomly combined to generate multiple parameter value combinations, where the parameter value combinations include one value of each model parameter.

[0066] In the present embodiment, the statistical method includes a moving average method, a moving median method, a weighted moving average method, and an exponentially weighted moving average method. When the statistical method is a moving average method, a moving median method, and a weighted moving average method, the model parameters include a step size, an upper error limit, and a lower error limit. When the statistical method is an exponentially weighted moving average method, the model parameters include a step size, an upper error limit, a lower error limit, and a weighting coefficient. Wherein, the step size is the number of serum potassium test data that need to be included when performing a real-time quality control. The upper error limit and the lower error limit are used to determine whether the quality control value obtained by performing a real-time quality control is out of control. When the quality control value is greater than the upper error limit or less than the lower error limit, the quality control value is out of control.

[0067] Specifically, according to the amount of data and the fluctuation characteristics, the appropriate step size, upper error limit and lower error limit are selected. First, the maximum acceptable step size is determined. For example, if the maximum step size has been determined to be 80, then the multiple values ​​of the step size can be 5, 10, 15, 20, ... 75, 80. The upper error limit and the lower error limit are determined according to the mean and standard deviation of the second training set. The multiple values ​​of the upper error limit can be mean + M * standard deviation, M = 0.5, 1, 1.5, 2, ..., 4.5, 5. The multiple values ​​of the lower error limit can be mean - M * standard deviation. The multiple values ​​of the weighting coefficient can be determined according to user needs. Optionally, the model parameters can also include an upper alarm limit and a lower alarm limit. The upper alarm limit and the lower alarm limit are used to determine whether the quality control value obtained by performing a real-time quality control needs to be paid attention to. When the quality control value is greater than the upper alarm limit or less than the lower alarm limit, the quality control value needs to be paid attention to and may be out of control later. The upper alarm limit is less than the upper error limit, and the lower alarm limit is greater than the lower error limit. The upper alarm limit and the lower alarm limit are determined based on the mean and standard deviation of the second training set. The multiple values ​​of the upper alarm limit can be mean + Q * standard deviation, Q = 0.5, 1, 1.5, 2, ..., 4.5, 5. The multiple values ​​of the lower alarm limit can be mean - Q * standard deviation.

[0068] (2) For each parameter value combination, a second initial model is determined based on the parameter value combination, and the second initial model is used to perform real-time quality control on the second training set to obtain the performance of the second initial model.

[0069] The performance includes a false positive rate, a false negative rate, and a timely rate. At this time, the second initial model is used to perform real-time quality control on the second training set to obtain the performance of the second initial model, specifically including: using the second initial model to perform real-time quality control on the second training set to obtain the false positive rate corresponding to the second initial model; based on the total allowable error, adding an error to each historical serum potassium test data in the second training set to obtain a second error training set; using the second initial model to perform real-time quality control on the second error training set to obtain the false negative rate and timely rate corresponding to the second initial model; and using the false positive rate, false negative rate, and timely rate corresponding to the second initial model as the performance of the second initial model.

[0070] (3) Select the second initial model with the best performance as the second model corresponding to the statistical method, and use the performance of the second initial model with the best performance as the performance of the second model to obtain the second model corresponding to each statistical method and the performance of each second model.

[0071] Selecting the second initial model with the best performance as the second model corresponding to the statistical method specifically includes: judging whether the absolute value of the difference between the false positive rate corresponding to the second initial model and the expected false positive rate is less than a preset difference; if so, determining the second initial model as the second post-screening initial model; calculating the sum of the false positive rate, false negative rate and timely rate corresponding to the second post-screening initial model, and selecting the second post-screening initial model with the smallest sum as the second model corresponding to the statistical method.

[0072] Similarly, the third training set is used as input and multiple statistical methods are used to perform modeling. The third model corresponding to each statistical method and the performance of each third model are obtained, specifically including:

[0073] (1) For each statistical method, multiple values ​​of each model parameter corresponding to the statistical method are determined, and the multiple values ​​of each model parameter are randomly combined to generate multiple parameter value combinations, where the parameter value combinations include one value of each model parameter.

[0074] In the present embodiment, the statistical method includes a moving average method, a moving median method, a weighted moving average method, and an exponentially weighted moving average method. When the statistical method is a moving average method, a moving median method, and a weighted moving average method, the model parameters include a step size, an upper error limit, and a lower error limit. When the statistical method is an exponentially weighted moving average method, the model parameters include a step size, an upper error limit, a lower error limit, and a weighting coefficient. Wherein, the step size is the number of serum potassium test data that need to be included when performing a real-time quality control. The upper error limit and the lower error limit are used to determine whether the quality control value obtained by performing a real-time quality control is out of control. When the quality control value is greater than the upper error limit or less than the lower error limit, the quality control value is out of control.

[0075] Specifically, according to the amount of data and the fluctuation characteristics, the appropriate step size, upper error limit and lower error limit are selected. First, the maximum acceptable step size is determined. For example, if the maximum step size has been determined to be 80, then the multiple values ​​of the step size can be 5, 10, 15, 20, ... 75, 80. The upper error limit and the lower error limit are determined according to the mean and standard deviation of the third training set. The multiple values ​​of the upper error limit can be mean + M * standard deviation, M = 0.5, 1, 1.5, 2, ..., 4.5, 5. The multiple values ​​of the lower error limit can be mean - M * standard deviation. The multiple values ​​of the weighting coefficient can be determined according to user needs. Optionally, the model parameters can also include an upper alarm limit and a lower alarm limit. The upper alarm limit and the lower alarm limit are used to determine whether the quality control value obtained by performing a real-time quality control needs to be paid attention to. When the quality control value is greater than the upper alarm limit or less than the lower alarm limit, the quality control value needs to be paid attention to and may be out of control later. The upper alarm limit is less than the upper error limit, and the lower alarm limit is greater than the lower error limit. The upper alarm limit and the lower alarm limit are determined based on the mean and standard deviation of the third training set. The multiple values ​​of the upper alarm limit can be mean + Q * standard deviation, Q = 0.5, 1, 1.5, 2, ..., 4.5, 5. The multiple values ​​of the lower alarm limit can be mean - Q * standard deviation.

[0076] (2) For each parameter value combination, a third initial model is determined based on the parameter value combination, and the third initial model is used to perform real-time quality control on the third training set to obtain the performance of the third initial model.

[0077] The performance includes the false positive rate, the false negative rate and the timely rate. At this time, the third initial model is used to perform real-time quality control on the third training set to obtain the performance of the third initial model, specifically including: using the third initial model to perform real-time quality control on the third training set to obtain the false positive rate corresponding to the third initial model; based on the total allowable error, adding an error to each historical serum potassium test data in the third training set to obtain the third error training set; using the third initial model to perform real-time quality control on the third error training set to obtain the false negative rate and timely rate corresponding to the third initial model; and taking the false positive rate, false negative rate and timely rate corresponding to the third initial model as the performance of the third initial model.

[0078] (3) Select the third initial model with the best performance as the third model corresponding to the statistical method, and use the performance of the third initial model with the best performance as the performance of the third model to obtain the third model corresponding to each statistical method and the performance of each third model.

[0079] Selecting the third initial model with the best performance as the third model corresponding to the statistical method specifically includes: judging whether the absolute value of the difference between the false positive rate corresponding to the third initial model and the expected false positive rate is less than a preset difference; if so, determining the third initial model as the third post-screening initial model; calculating the sum of the false positive rate, false negative rate and timely rate corresponding to the third post-screening initial model, and selecting the third post-screening initial model with the smallest sum as the third model corresponding to the statistical method.

[0080] Among them, selecting the first model with the best performance as the first real-time quality control model specifically includes: judging whether the absolute value of the difference between the false positive rate corresponding to the first model and the expected false positive rate is less than the preset difference; if so, determining the first model as the first post-screening model; calculating the sum of the false positive rate, false negative rate and timely rate corresponding to the first post-screening model, and selecting the first post-screening model with the smallest sum as the first real-time quality control model corresponding to the statistical method.

[0081] Similarly, the second model with the best performance is selected as the second real-time quality control model, which specifically includes: judging whether the absolute value of the difference between the false positive rate corresponding to the second model and the expected false positive rate is less than the preset difference; if so, determining the second model as the second post-screening model; calculating the sum of the false positive rate, false negative rate and timely rate corresponding to the second post-screening model, and selecting the second post-screening model with the smallest sum as the second real-time quality control model corresponding to the statistical method.

[0082] Similarly, the third model with the best performance is selected as the third real-time quality control model, specifically including: judging whether the absolute value of the difference between the false positive rate corresponding to the third model and the expected false positive rate is less than the preset difference; if so, determining the third model as the third post-screening model; calculating the sum of the false positive rate, false negative rate and timeliness rate corresponding to the third post-screening model, and selecting the third post-screening model with the smallest sum as the third real-time quality control model corresponding to the statistical method.

[0083] In this embodiment, the appropriate statistical methods selected based on the characteristics of each group of pre-processed training data are as follows: (1) The dialysis group and the non-dialysis group: the exponentially weighted moving average method is used because it is more sensitive to the latest data, suitable for processing data with large fluctuations, and can quickly respond to changes in serum potassium; (2) The moving median method is used for the physical examination group because it has a strong ability to filter outliers and is suitable for processing relatively stable data to filter out outliers and noise information.

[0084] Step 4: Model verification.

[0085] Step 4 uses the historical serum potassium test data of the different types of people in another time period as validation data set, verifies the three PBRTQC models built in step 3. Use the historical serum potassium test data of the different types of people in another time period as validation data set, ensure that the time distribution of validation data set is different from that of training data set, equally validation data set is divided into three groups of validation data, for each group of validation data, the PBRTQC model corresponding to the group of validation data is used to carry out real-time quality control on the group of validation data, obtain false positive rate, then introduce error (such as 1 / 2TEA and 1 / 3TEA) in the group of validation data, the PBRTQC model corresponding to the group of validation data is used to carry out real-time quality control on the group of validation data after adding error, obtain timely rate and detection rate, and use other groups of validation data to verify the applicability of the PBRTQC model corresponding to the group of validation data, assess whether false positive rate is significantly reduced. The probability of false alarm of model detection result error is evaluated by false positive rate, the speed of error is detected by timely rate evaluation model, the ability of error is correctly identified by detection rate evaluation model, validation result confirms that the crowd of corresponding type is detected using corresponding model, and false positive rate is the lowest. Detection rate APed = the average of the ratios of the number of quality control values ​​that exceed the upper and lower error limits after adding the error to the total number of quality control values ​​with added errors.

[0086] At this time, in this embodiment, after obtaining the first real-time quality control model, the second real-time quality control model, and the third real-time quality control model, the method for establishing a serum potassium real-time quality control model based on pre-classification in this embodiment further includes:

[0087] (1) Obtain a first validation set, a second validation set, and a third validation set, wherein the first validation set includes multiple historical serum potassium test data of dialysis patients in the second historical period, the second validation set includes multiple historical serum potassium test data of healthy subjects in the second historical period, and the third validation set includes multiple historical serum potassium test data of non-dialysis patients in the second historical period. The length of the second historical period can be determined according to user needs, and the second historical period is different from the first historical period.

[0088] (2) The first validation set is divided into multiple validation subsets, and different validation subsets correspond to different seasons.

[0089] (3) Using the first real-time quality control model to perform real-time quality control on the first validation set and each validation subset, respectively, to obtain the false positive rate, timely rate and detection rate corresponding to the first real-time quality control model in the first validation set and each validation subset, respectively, to verify the first real-time quality control model.

[0090] (4) Using the second real-time quality control model, real-time quality control is performed on the second validation set and each validation subset respectively, and the false positive rate, timely rate and detection rate corresponding to the second real-time quality control model in the second validation set and each validation subset are obtained to verify the second real-time quality control model.

[0091] (5) Using the third real-time quality control model, real-time quality control is performed on the third validation set and each validation subset respectively, and the false positive rate, timely rate and detection rate corresponding to the third real-time quality control model in the third validation set and each validation subset are obtained to verify the third real-time quality control model.

[0092] Step 5: Implementation and monitoring.

[0093] The PBRTQC model validated in step 4 is applied to the laboratory testing system to enable real-time monitoring of serum potassium test data. Specifically, the constructed PBRTQC model is applied to the laboratory testing system to enable real-time monitoring of serum potassium test data. Step 5 applies the method to real-world scenarios to ensure the long-term accuracy and reliability of the laboratory testing system.

[0094] This example removes dialysis patients from all patients and establishes PBRTQC models for dialysis patients, healthy subjects, and non-dialysis patients, respectively. The false positive rates of the PBRTQC models for different populations are compared. It is verified that because dialysis patients' serum potassium levels are prone to high and low values, the PBRTQC model generates many false positive alarms. Establishing separate models allows for more accurate real-time monitoring of patient serum potassium test data and instruments. By pre-classifying the data and constructing separate PBRTQC models, this example effectively reduces PBRTQC false positive alarms in serum potassium testing caused by the presence of dialysis patients, eliminates this special population of dialysis patients, and improves the accuracy and reliability of the laboratory testing system.

[0095] The following is an example to introduce the method for establishing a real-time quality control model for serum potassium based on pre-classification in this embodiment:

[0096] A total of 362,077 historical serum potassium test data from a medical center from September 2023 to September 2024 were collected. All historical serum potassium test data were divided into dialysis group, physical examination group and non-dialysis group. The data from September 2023 to February 2024 were selected as the training data set (3,217 cases in the dialysis group, 7,339 cases in the physical examination group, and 153,565 cases in the non-dialysis group) to build PBRTQC models for the dialysis group, physical examination group and non-dialysis group respectively. Three validation sets were created to validate the three PBRTQC models. Validation set 1 contained data from the dialysis group from March to September 2024 (3,957 cases), validation set 2 contained data from the non-dialysis group from March to April 2024 (42,622 cases), and validation set 3 contained data from the physical examination group from May to June 2024 (1,345 cases). In order to consider the potassium ion (K + ) levels, validation set 1 was further divided into three subgroups according to the time sequence: subgroup a (1224 cases from March to April), subgroup b (1234 cases from May to June), and subgroup c (1499 cases from July to September).

[0097] Historical serum potassium test data were all tested using the Roche COBAS 702 biochemical analyzer. Routine quality control was performed before testing. Quality control products and calibrators were all from Roche. Patient testing could only be performed after passing the internal quality control. All historical serum potassium test data were collected from the laboratory information system. According to relevant regulations, all symbols reflecting patient information were deleted during the data collection and collation process to ensure the confidentiality of the entire experimental process.

[0098] The collected historical serum potassium data were divided into three distinct groups: the dialysis group, the physical examination group, and the non-dialysis group. This classification was crucial to ensure that the unique physiological characteristics of each group were adequately addressed in subsequent analyses. Four statistical methods—moving mean (MM), moving median (MA), weighted moving average (WMA), and exponential weighted moving average (EWMA)—were used to construct a PBRTQC model for each group. The method first set the total allowable error and expected false positive rate. The false positive rate was then calculated for various model parameters. The error was then added to calculate the false negative rate and timely rate. The optimal model parameters were determined based on the false positive rate, false negative rate, and timely rate, resulting in the PBRTQC model.

[0099] Before establishing the PBRTQC model, the three data sets were first processed to exclude non-serum samples. The final data set included 3217 cases in the dialysis group, 7339 cases in the physical examination group, and 153,565 cases in the non-dialysis group. Statistical analysis revealed statistically significant differences among the three groups (p < 0.001). The final parameters for the PBRTQC model for the three data sets are shown in Table 1. The cutoff ranges for the dialysis group were 2.22–8.04 mmol / L, the cutoff range for the non-dialysis group was 2.76–5.61 mmol / L, and the cutoff range for the physical examination group was 2.97–5.25 mmol / L. Based on the minimum sum of the FPR, FNR, and ANP, the EWMA method was used for modeling the dialysis and non-dialysis groups, and the MM method was used for modeling the physical examination group. The target values ​​(i.e., means) for the three data sets were similar: 4.04 mmol / L for both the dialysis and physical examination groups, and 4.17 mmol / L for the non-dialysis group. The three groups selected different step lengths, with 95 cases in the dialysis group, 25 cases in the non-dialysis group, and 35 cases in the physical examination group.

[0100] Table 1 Final parameters of the training set

[0101]

[0102]

[0103] Note: In Table 1, * indicates the final selected method.

[0104] Validation sets 1, 2, and 3 were used to validate the PBRTQC models of the dialysis group, physical examination group, and non-dialysis group. In this embodiment, validation set 1 was used to validate the PBRTQC models of the dialysis group, physical examination group, and non-dialysis group; validation set 2 was used to validate the PBRTQC model of the non-dialysis group; and validation set 3 was used to validate the PBRTQC model of the physical examination group. Before validation, validation set 1 was divided into three groups according to time sequence to exclude K + Affected by cyclical or seasonal trends, each set of data was verified in the positive and negative directions by adding 1 / 2TEA and 1 / 3TEA, respectively. According to the health industry standard of the People's Republic of China (WS / T403-2012), TEA was set at 6%. ANPed, Aped, and MNPed were selected to evaluate the PBRTQC model, where ANPed is the average number of affected patient results before error detection, Aped is the detection rate, and MNPed is the median number of affected patient results before error detection. The smaller the ANPed and MNPed, the more sensitive the PBRTQC model, and the larger the Aped, the better the detection rate of the PBRTQC model.

[0105] After selecting the PBRTQC model, the validation set 1 is used to validate the established PBRTQC model, considering the different time effects on K +The effect of the validation set 1 was divided into three groups (i.e., a, b, and c) in chronological order, and the validation set 1 was validated separately. At the same time, 1 / 2TEA and 1 / 3TEA were introduced in the positive and negative directions, respectively, to calculate the timely rate, detection rate, and false positive rate. The results are shown in Table 2. In the physical examination group and the dialysis group, there was no significant difference in the false positive rate of the validation set data of groups a, b, and c (p>0.05). However, in the non-dialysis group, there was a significant difference in the false positive rate of the validation set data of groups a, b, and c (p<0.05), indicating that the serum K + The results of the present invention change significantly over time. This example found that the mean serum potassium values ​​from April to June and from July to September were higher than those from October to December and from January to March, and the differences were statistically significant (p < 0.001). Figure 4 The false positive rates of the three models in validation set 1 were 69.26% for the physical examination group, 1.143% for the dialysis group, and 35.68% for the non-dialysis group (see Table 2 and Figure 4 )(p<0.05), indicating that the dialysis group K + The results were significantly different from those of the other two groups. The timely rate of the dialysis group (1 / 2TEA, positive timely rate: 307.30, negative timely rate: 795.20) was significantly higher than that of the physical examination group (1 / 2TEA, positive timely rate: 10.57, negative timely rate: 4.67) and the non-dialysis group (1 / 2TEA, positive timely rate: 24.57, negative timely rate: 29.57), while the detection rate of the dialysis group (1 / 2TEA, positive detection rate: 2.83%, negative detection rate: 0.67%) was significantly lower than that of the physical examination group (1 / 2TEA, positive detection rate: 41.47%, negative detection rate: 45.11%) and the non-dialysis group (1 / 2TEA, positive detection rate: 16.00%, negative detection rate: 18.00%). In validation sets 2 and 3, the false positive rates of the non-dialysis group and the physical examination group were 1.906% and 2.83%, respectively, which indicates that pre-classification of dialysis patients can significantly reduce the occurrence of false positive results. In addition, the serum potassium test results of the non-dialysis group showed obvious seasonal changes. In this embodiment, 1 / 2TEA and 1 / 3TEA were introduced in the positive and negative directions, respectively. When the first set of data was used to verify the dialysis parameters, the negative timely rate and negative detection rate were not detected when 1 / 2TEA was introduced, and the values ​​of the remaining data groups were close.

[0106] Table 2 Validation of the PBRTQC model for the physical examination, dialysis, and non-dialysis groups using validation sets 1, 2, and 3, respectively

[0107]

[0108]

[0109] Note: In Table 2, NA stands for not detected.

[0110] The validation results of validation set 1 showed that the false positive rate in the dialysis group was significantly lower than that in the physical examination group and the non-dialysis group (p < 0.05). The validation results of validation set 2 showed that the false positive rate in the non-dialysis group was lower than that in validation set 1. This example used a method of pre-classifying patients' serum potassium to establish PBRTQC models using different statistical methods. The three models were then validated using dialysis data. By analyzing parameters such as false positive rate, timely rate, and detection rate, it was confirmed that serum potassium in dialysis patients interfered with the false positive effects of the PBRTQC model, emphasizing the importance of pre-classifying dialysis patients.

[0111] Different data distributions or project and population characteristics require different statistical methods. In this embodiment, three PBRTQC models were established for three different populations. Among them, the dialysis group and the non-dialysis group ultimately chose the EWMA method for modeling, while the physical examination group adopted the moving median method for modeling. This is consistent with previous research conclusions. The physical examination group selected the moving median method as the optimal method due to the relatively stable serum potassium value. The moving median method has good robustness and can better filter out outliers and noise information. It is suitable for processing relatively stable biomarker data. The EWMA method can quickly respond to changes in serum potassium by giving greater weight to the latest results, and can adjust offset sensitivity. It has advantages in detecting small quality-related changes with lower false positive alarms. The EWMA method attempts to accumulate all past small fluctuations to make them appear to have larger fluctuations, thereby increasing the sensitivity of the control chart to small drifts and being able to detect systematic errors more sensitively and early.

[0112] In the detection of serum potassium, dialysis patients show unique biological characteristics and are affected by multiple factors. The detection value of serum potassium is affected by multiple factors such as the patient's dietary habits, dialysis regimen, dialysis time, and individual physiological state. For example, when patients consume a high-potassium diet or have renal insufficiency, serum potassium concentration often increases. In addition, the removal efficiency of blood potassium during dialysis is also an important influencing factor. Certain dialysis membrane materials and dialysate components can lead to different serum potassium removal effects. The serum potassium level of dialysis patients usually shows the characteristics of being high before dialysis and low after dialysis. Before dialysis, due to the loss of renal function in the patient's body, the excess potassium ions in the body are not effectively excreted, resulting in an increase in serum potassium concentration. After the dialysis process, potassium ions will be effectively removed, so that the serum potassium level after dialysis is significantly reduced. In addition, studies have shown that the serum potassium level of dialysis patients is also affected by seasonal changes, and the fluctuation of electrolytes varies significantly in different seasons. In this example, significant differences in serum potassium levels were observed between the dialysis group, the physical examination group, and the non-dialysis group. In addition, when verifying the data of the non-dialysis group, it was divided into three groups (Group 1: March to April, Group 2: May to June, Group 3: July to August). The results showed that there were significant differences among the three groups (p<0.05), among which the false positive rate of the first group of data (March to April) was the highest, which may be related to seasonal changes. Due to the seasonal alternation in spring, changes in patients' living habits and dietary structure may lead to significant changes in serum potassium levels. Figure 4 It can be seen that in this embodiment, the serum potassium level varies with the seasons, and the average value from April to September is higher than that from October to March. This is similar to the results of a Swiss study, in which the serum potassium level reached the lowest point in October and the highest point in April. This may be because April to September is the summer and autumn seasons, during which there are more potassium-containing vegetables and fruits, which may lead to differences in potassium intake in the diet.

[0113] The importance of pre-classifying patients before establishing the PBRTQC model is fully reflected in this embodiment. The results show that when the data of the dialysis group in the second half of the year are used as the validation set, the false positive rates of the three groups are 69.257% for the physical examination group, 1.143% for the dialysis group and 35.675% for the non-dialysis group. In response to this result, part of the non-dialysis data in the second half of the year was also used to verify the non-dialysis group, and the false positive rate was 1.906%, which is in sharp contrast to the false positive rate of 35.675% obtained in the dialysis group validation set. This significant difference shows that pre-classification can not only effectively reduce the false positive rate, but also the false positive rates of 69.257% and 35.675% in the non-dialysis group and the physical examination group, which show large fluctuations, indicating that the serum potassium levels of these patients fluctuate greatly. Therefore, the implementation of pre-classification can greatly improve the accuracy of monitoring and reduce unnecessary interference. Although the false positive rate was low (1.143%) in the dialysis group, the median of the affected patients' results before error detection was higher, indicating insufficient sensitivity. This phenomenon may be due to the fact that the K + The fluctuation before and after dialysis is large, and it is easy to have extremely high and low values. Therefore, when establishing the PBRTQC model, a wider cutoff range was selected to reduce the false positive rate. Although 1 / 2TEA and 1 / 3TEA have been introduced, TEA is only 6%, which is not enough to make up for the K of dialysis patients. + The fluctuations resulted in higher MNped and ANped. This issue may be resolved by modeling pre-dialysis and post-dialysis samples separately.

[0114] When further analyzing the trend of false positives in the dialysis and non-dialysis groups, it was noted that the false positives had similar time intervals and showed a consistent pattern, e.g. Figure 5 As shown, this is primarily due to serum potassium being measured before and after dialysis in dialysis patients. Predialysis serum potassium values ​​are typically significantly higher than the normal range, while postdialysis serum potassium values ​​are often lower than the normal range. Dialysis patients undergo treatment on fixed dialysis days, resulting in samples with both high and low values ​​within specific time periods.

[0115] By pre-classifying patient data and constructing a PBRTQC model for each, the team significantly reduced the false positive rate in serum potassium testing and improved the real-time monitoring accuracy of the laboratory testing system. This approach is particularly useful for eliminating false positive alarms in the PBRTQC program caused by results from dialysis patients, whose serum potassium levels fluctuate greatly and are prone to false alarms.

[0116] The present application also provides an application scenario, which applies the above-mentioned method for establishing a real-time quality control model for serum potassium based on pre-classification. Specifically, the method for establishing a real-time quality control model for serum potassium based on pre-classification provided in the present embodiment can be applied in a real-time quality control scenario. The real-time quality control scenario includes an establishment link and an application link. The establishment link is used to respectively establish a first real-time quality control model for dialysis patients, a second real-time quality control model for healthy physical examinees, and a third real-time quality control model for non-dialysis patients. The application link is used to perform real-time quality control on the serum potassium test data of dialysis patients using the first real-time quality control model, perform real-time quality control on the serum potassium test data of healthy physical examinees using the second real-time quality control model, and perform real-time quality control on the serum potassium test data of non-dialysis patients using the third real-time quality control model. The method for establishing a real-time quality control model for serum potassium based on pre-classification provided in the present embodiment belongs to the establishment link.

[0117] Example 2

[0118] This embodiment provides an application method of a real-time quality control model for serum potassium based on pre-classification, such as Figure 6 Shown, including:

[0119] Step T1: Perform real-time quality control on serum potassium test data of dialysis patients using a first real-time quality control model; the first real-time quality control model is a first real-time quality control model obtained by using the method for establishing a real-time quality control model of serum potassium based on pre-classification described in Example 1.

[0120] Step T2: Perform real-time quality control on the serum potassium test data of the healthy subjects using a second real-time quality control model; the second real-time quality control model is a second real-time quality control model obtained by the method for establishing a serum potassium real-time quality control model based on pre-classification described in Example 1.

[0121] Step T3, using a third real-time quality control model to perform real-time quality control on the serum potassium test data of non-dialysis patients; the third real-time quality control model is a third real-time quality control model obtained by using the method for establishing a serum potassium real-time quality control model based on pre-classification described in Example 1.

[0122] Example 3

[0123] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 7As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for establishing a real-time quality control model of serum potassium based on pre-classification or a method for applying a real-time quality control model of serum potassium based on pre-classification is implemented.

[0124] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0125] In an exemplary embodiment, a computer device is provided, comprising a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the method for establishing a real-time quality control model of serum potassium based on pre-classification in Example 1 or the method for applying a real-time quality control model of serum potassium based on pre-classification in Example 2 is implemented.

[0126] Example 4

[0127] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, which, when executed by a processor, implements the method for establishing a real-time quality control model for serum potassium based on pre-classification in Example 1 or the method for applying a real-time quality control model for serum potassium based on pre-classification in Example 2.

[0128] Example 5

[0129] In an exemplary embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the method for establishing a real-time quality control model for serum potassium based on pre-classification in Example 1 or the method for applying a real-time quality control model for serum potassium based on pre-classification in Example 2.

[0130] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0131] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0132] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for establishing a real-time quality control model for serum potassium based on pre-classification, characterized in that: include: Obtain a first training set, a second training set, and a third training set; the first training set includes multiple historical serum potassium test data of dialysis patients in a first historical period, the second training set includes multiple historical serum potassium test data of healthy subjects in the first historical period, and the third training set includes multiple historical serum potassium test data of non-dialysis patients in the first historical period; Using the first training set as input, modeling is performed using multiple statistical methods to obtain a first model corresponding to each of the statistical methods and the performance of each of the first models; Using the second training set as input, modeling is performed using multiple statistical methods to obtain a second model corresponding to each of the statistical methods and the performance of each of the second models; Using the third training set as input, modeling is performed using multiple statistical methods to obtain a third model corresponding to each statistical method and the performance of each third model; The first model with the best performance is selected as the first real-time quality control model, the second model with the best performance is selected as the second real-time quality control model, and the third model with the best performance is selected as the third real-time quality control model; the first real-time quality control model is used to perform real-time quality control on the serum potassium detection data of dialysis patients, the second real-time quality control model is used to perform real-time quality control on the serum potassium detection data of healthy physical examination subjects, and the third real-time quality control model is used to perform real-time quality control on the serum potassium detection data of non-dialysis patients.

2. The method for establishing a serum potassium real-time quality control model based on pre-classification according to claim 1, wherein The method includes: performing data cleaning on the first training set to obtain a first cleaned training set; and performing modeling on the first cleaned training set to obtain a first model corresponding to each statistical method and the performance of each first model. The method includes removing outliers, wherein the outliers are historical serum potassium test data obtained by testing non-serum specimens, historical serum potassium test data that is not within a reasonable range, and historical serum potassium test data that is not within a first cutoff range. The first cutoff range is determined based on the data distribution of the historical serum potassium test data in the first training set.

3. The method for establishing a serum potassium real-time quality control model based on pre-classification according to claim 1, wherein Using the first training set as input, modeling is performed using multiple statistical methods to obtain a first model corresponding to each of the statistical methods and the performance of each of the first models, specifically including: For each of the statistical methods, determining multiple values ​​of each model parameter corresponding to the statistical method, and randomly combining the multiple values ​​of each of the model parameters to generate multiple parameter value combinations; the parameter value combinations include one value of each of the model parameters; For each of the parameter value combinations, determining a first initial model based on the parameter value combination, and performing real-time quality control on the first training set using the first initial model to obtain performance of the first initial model; The first initial model with the best performance is selected as the first model corresponding to the statistical method, and the performance of the first initial model with the best performance is used as the performance of the first model to obtain the first model corresponding to each of the statistical methods and the performance of each of the first models.

4. The method for establishing a serum potassium real-time quality control model based on pre-classification according to claim 3, wherein The statistical methods include a moving average method, a moving median method, a weighted moving average method, and an exponentially weighted moving average method. When the statistical methods are the moving average method, the moving median method, and the weighted moving average method, the model parameters include a step size, an upper error limit, and a lower error limit. When the statistical method is the exponentially weighted moving average method, the model parameters include a step size, an upper error limit, a lower error limit, and a weighting coefficient. The step size is the number of serum potassium test data that needs to be included when performing a real-time quality control. The upper error limit and the lower error limit are used to determine whether the quality control value obtained by performing a real-time quality control is out of control. When the quality control value is greater than the upper error limit or less than the lower error limit, the quality control value is out of control.

5. The method for establishing a serum potassium real-time quality control model based on pre-classification according to claim 3, wherein The performance includes a false positive rate, a false negative rate, and a timely rate. At this time, the first initial model is used to perform real-time quality control on the first training set to obtain the performance of the first initial model, which specifically includes: Performing real-time quality control on the first training set using the first initial model to obtain a false positive rate corresponding to the first initial model; Based on the total allowable error, adding an error to each historical serum potassium test data in the first training set to obtain a first error training set; Using the first initial model to perform real-time quality control on the first error training set, to obtain a false negative rate and a timely rate corresponding to the first initial model; The false positive rate, false negative rate and timely rate corresponding to the first initial model are used as the performance of the first initial model.

6. The method for establishing a serum potassium real-time quality control model based on pre-classification according to claim 5, wherein Selecting the first initial model with the best performance as the first model corresponding to the statistical method specifically includes: Determine whether the absolute value of the difference between the false positive rate corresponding to the first initial model and the expected false positive rate is less than a preset difference; If yes, the first initial model is determined as the first post-screening initial model; The sum of the false positive rate, false negative rate and timely rate corresponding to the first initial model after screening is calculated, and the first initial model after screening with the smallest sum is selected as the first model corresponding to the statistical method.

7. A method for applying a real-time quality control model for serum potassium based on pre-classification, characterized in that: include: Performing real-time quality control on serum potassium test data of dialysis patients using a first real-time quality control model; the first real-time quality control model is a first real-time quality control model obtained by using the method for establishing a real-time quality control model of serum potassium based on pre-classification according to any one of claims 1 to 6; Performing real-time quality control on the serum potassium test data of healthy subjects using a second real-time quality control model; the second real-time quality control model is a second real-time quality control model obtained by using the method for establishing a serum potassium real-time quality control model based on pre-classification according to any one of claims 1 to 6; Real-time quality control is performed on serum potassium detection data of non-dialysis patients using a third real-time quality control model; the third real-time quality control model is a third real-time quality control model obtained by using the method for establishing a serum potassium real-time quality control model based on preclassification according to any one of claims 1 to 6.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for establishing a real-time quality control model of serum potassium based on pre-classification according to any one of claims 1 to 6 or the method for applying a real-time quality control model of serum potassium based on pre-classification according to claim 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for establishing a real-time quality control model of serum potassium based on pre-classification according to any one of claims 1 to 6 or the method for applying a real-time quality control model of serum potassium based on pre-classification according to claim 7 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method for establishing a real-time quality control model of serum potassium based on pre-classification according to any one of claims 1 to 6 or the method for applying a real-time quality control model of serum potassium based on pre-classification according to claim 7 is implemented.

Images