Single-bottle reagent quality dynamic monitoring method and storage medium

Abstract

A method and storage medium for dynamic monitoring of single-bottle reagent quality are disclosed, comprising: acquiring the test result of the current reagent on the current sample and multi-source detection background data about the current sample, wherein the multi-source detection background data includes one or more of patient characteristic data, testing hospital, and testing date; obtaining the feature vector of the current sample based on the multi-source detection background data and test result; obtaining the positive probability of the current sample based on the feature vector of the current sample; acquiring historical samples before the current sample and their positive probabilities and test results to calculate the abnormal probability of the current reagent, thereby determining whether the quality of the current reagent is abnormal. By fusing the test results with multi-source detection background data, the method adapts to different patient populations, reducing the false alarm rate of single-bottle reagent abnormalities; and by using the positive probabilities of the current sample and historical samples, it achieves dynamic generation of hospital-specific thresholds, thereby realizing real-time quality control of each bottle of reagent.

Description

Technical Field

[0001] This invention relates to the field of medical testing technology, specifically to a method and storage medium for dynamic monitoring of the quality of a single-bottle reagent. Background Technology

[0002] Traditional reagent quality monitoring methods mainly rely on fixed thresholds for judgment, which are difficult to adapt to the differences in populations in different hospitals (such as differences in medical decision-making levels between the north and south) and the impact of seasonal fluctuations (such as the increase in batch positive results for corresponding test items during the peak season of respiratory infections in winter).

[0003] In addition, existing analytical methods are relatively simple, relying on deviations in internal quality control samples to determine whether reagents are abnormal. However, due to limitations in cost, resources, and efficiency, laboratories often cannot perform quality control on every bottle of reagent, making it difficult to detect abnormalities in individual bottles.

[0004] Meanwhile, the adjustment of reagent quality monitoring thresholds is significantly delayed, mainly relying on manual retrospective analysis. The cycle from identifying anomalies to updating thresholds usually exceeds 48 hours, during which unreliable test results are continuously generated, thus affecting the accuracy of clinical judgment. Summary of the Invention

[0005] The main technical problem addressed by this invention is how to overcome the technical bottleneck that reagent quality monitoring methods based on static thresholds cannot adapt to differences between patient populations and different regions, and how to monitor the quality of individual reagent bottles in real time to avoid producing unreliable test results.

[0006] According to the first aspect, one embodiment provides a method for obtaining the detection result corresponding to the current sample after the current reagent performs the corresponding detection item on the current sample;

[0007] Obtain the test results for the current sample after performing the corresponding test on the current sample with the current reagent;

[0008] Obtain multi-source detection background data for the current sample, wherein the multi-source detection background data includes one or more of patient characteristic data, testing hospital, and testing date;

[0009] Based on the multi-source detection background data and detection results of the current sample, the feature vector of the current sample is obtained;

[0010] The positive probability of the current sample is obtained based on the feature vector of the current sample;

[0011] Obtain historical samples preceding the current sample, and obtain the positive probability and detection results of the historical samples;

[0012] The abnormal probability of the current reagent is calculated based on the positive probability and test results of the current sample and the historical samples.

[0013] Determine whether there is an abnormality in the quality of the current reagent based on the probability of abnormality of the current reagent.

[0014] In some embodiments, the patient characteristic data includes one or more of the patient's age group, gender, disease label, and patient's geographic location.

[0015] In some embodiments, obtaining the feature vector of the current sample based on the multi-source detection background data and detection results includes:

[0016] Based on the multi-source detection background data and detection results of the current sample, feature encoding is performed to convert discrete classification features into numerical features, thereby obtaining the feature vector of the current sample.

[0017] In some embodiments, obtaining the positive probability of the current sample based on the feature vector of the current sample includes:

[0018] The feature vector of the current sample is input into the probability prediction model of the test item corresponding to the current reagent to obtain the positive probability of the current sample; the probability prediction model of the test item corresponding to the current reagent is trained by the following method:

[0019] Obtain a multi-source dataset, which contains multi-source detection background data for each sample and the detection results obtained after each sample has been tested for corresponding detection items using multiple reagents;

[0020] For the current reagent-corresponding test item: extract the test result corresponding to each sample for the test item based on the multi-source dataset; obtain the feature vector of each sample based on the test result corresponding to each sample for the test item and the multi-source test background data; input the feature vector of each sample into a preset network to obtain the positive probability corresponding to the sample; calculate the loss function based on the positive probability corresponding to each sample and the test result; train the preset network based on the loss function until the loss function converges to obtain the probability prediction model corresponding to the test item.

[0021] In some embodiments, the method further includes: acquiring device log data, which is used to determine whether the device is under abnormal conditions.

[0022] In some embodiments, after acquiring the multi-source dataset and before extracting the detection result corresponding to each sample for the detection item based on the multi-source dataset, the method further includes:

[0023] The multi-source dataset is preprocessed, including outlier removal and data association.

[0024] The outlier removal is used to remove test results in the multi-source dataset that are under abnormal equipment conditions based on equipment log data; the data association is used to associate the test results corresponding to each reagent with the multi-source test background data of the corresponding sample.

[0025] In some embodiments, obtaining historical samples prior to the current sample includes obtaining a preset number of historical samples prior to the current sample, wherein the historical samples include samples that were tested prior to the current sample by the current reagent and / or reagents of the same type as the current reagent.

[0026] In some embodiments, calculating the abnormal probability of the current reagent based on the positive probability and test results corresponding to the current sample and the historical samples includes:

[0027] The number of positive samples is determined based on the test results corresponding to the current samples and the historical samples.

[0028] The significance probability is calculated based on the number of positive samples and the positive probabilities corresponding to the current sample and the historical samples; the significance probability is used as the anomaly probability of the current reagent.

[0029] In some embodiments, determining whether the quality of the current reagent is abnormal based on the probability of abnormality of the current reagent includes:

[0030] The current reagent quality is assessed based on its anomaly probability and a preset significance level to determine whether there is an anomaly.

[0031] If an anomaly is detected, a single-bottle reagent anomaly alarm will be triggered; otherwise, the test results for the next sample using the current reagent will continue to be obtained.

[0032] According to the second aspect, one embodiment provides a storage medium storing a computer program that can be executed by a processor to implement the above-described method for dynamic monitoring of the quality of a single-bottle reagent.

[0033] According to the above embodiment, a method and storage medium for dynamic monitoring of single-bottle reagent quality are provided. By acquiring the test results of each sample after testing with a single-bottle reagent and the corresponding multi-source test background data, and based on a hospital-localized probability detection model, the positive probability of each sample under normal conditions, meeting the conditions of patient characteristics, region, and testing season, is obtained. Then, based on the positive probability of the current sample and historical samples and the test results, the significance probability under the assumption of normal reagent quality is calculated in real time, realizing the real-time adjustment of the dynamic P-value of single-bottle reagent, thereby enabling real-time monitoring of the current reagent quality. By fusing test results with multi-source background data, the assessment of reagent quality becomes more realistic, overcoming the technical bottleneck of traditional static threshold methods that cannot adapt to differences in patient populations, regions, and seasons, and reducing the false alarm rate of single-bottle abnormalities. By predicting the positive probability of the current sample under normal reagent quality after each test result obtained with the current reagent, and combining the positive probability of historical samples with test results, hospital-specific thresholds are dynamically generated, thereby automatically monitoring the quality of the current reagent in real time without manual intervention. This avoids unreliable test results caused by the lag in adjusting reagent quality monitoring thresholds in traditional methods, while still ensuring quality control for each bottle of reagent. Attached Figure Description

[0034] Figure 1 This is a flowchart of a method for dynamic monitoring of the quality of a single bottle of reagent;

[0035] Figure 2 This is a flowchart of the training process for the probability prediction model of each detection item. Detailed Implementation

[0036] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of this application. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to this application are not shown or described in the specification. This is to avoid obscuring the core parts of this application with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0037] Furthermore, the features, operations, or characteristics described in the specification can be combined in any suitable manner to form various embodiments. At the same time, the steps or actions in the method description can be rearranged or adjusted in a manner obvious to those skilled in the art. Therefore, the various orders in the specification and drawings are only for the clear description of a particular embodiment and do not imply a necessary order, unless otherwise stated that a particular order must be followed.

[0038] The serial numbers assigned to components in this document, such as "first" and "second," are used only to distinguish the described objects and have no sequential or technical meaning. The terms "connection" and "linkage" used in this application, unless otherwise specified, include both direct and indirect connections (linkages).

[0039] The disease spectrum describes the distribution of various diseases (or health problems) in a specific population, period, and region, according to indicators such as frequency of occurrence, severity, harm, and socioeconomic burden.

[0040] Internal quality control deviation refers to the comparison between the measured results of quality control samples of known concentrations and the expected target value (or allowable range) when tested using a specific detection system in the same laboratory. The stability of the reagent is judged based on whether the deviation between the two is within an acceptable range. However, the method of judging reagent quality by internal quality control deviation is costly and has low detection efficiency. It often cannot achieve full quality control, that is, it is impossible to perform quality control on every bottle of reagent, making it difficult to detect abnormalities in individual bottles of reagent.

[0041] In existing fixed threshold algorithms, during the testing process using a reagent bottle for a specific project, if the reagent is tested multiple times consecutively (N... total If the test result for a reagent is positive, it is considered that the reagent may be abnormal. Further analysis of the test results for this reagent and similar reagents is needed to determine if the reagent is indeed abnormal. First, the cumulative positive probability P1 (number of positive tests / total number of tests) for the current reagent is obtained. Then, combined with the overall positive rate of this test in the past and the distribution range of the single-bottle positive rate for each bottle of reagent in the past test, a first reference probability P2 (overall positive rate) and a second reference probability P3 (maximum single-bottle positive rate within the distribution range) are determined. Based on the cumulative positive probability of the current reagent and the first and second reference probabilities, it is determined whether the current reagent is abnormal. For example, when P1 ≥ N... total ×P2 or P1 ≥ N total When ×P3 is reached, it is considered that there is an abnormality in the current reagent and a single reagent bottle abnormality alarm is triggered to provide a notification.

[0042] However, the distribution of the same disease in the population is by no means uniform. Its incidence is affected by a combination of factors such as age, region, and season. For example, the elderly population is significantly more susceptible to specific diseases due to immune aging and comorbid chronic diseases. Differences in lifestyle and climate between the north and south lead to differences in the level of medical decision-making in different regions (i.e., localized differences), which in turn leads to large differences in the test results of the same indicator (such as blood lipids). Winter is the peak season for respiratory infections, and corresponding test items are prone to batch-to-batch positive results. Therefore, the existing methods only judge whether the reagent is abnormal based on the test results, without considering factors such as the age, region, and season of the corresponding patient, which is prone to false alarms of single-bottle abnormalities.

[0043] This invention relates to a method that associates each reagent bottle with multi-source background data for corresponding samples, including patient characteristics, testing hospitals, and testing dates. This data is then combined with the current reagent's test results to perform multi-dimensional data fusion, resulting in a feature vector for the current sample. A machine learning model is trained locally to predict the positive probability of the corresponding patient under the assumption of normal reagent quality. The number of positive samples detected by the reagent is then determined based on the current sample and historical samples. Based on the positive probabilities of these samples, the probability of at least one positive sample under normal reagent quality is calculated using a Poisson-Binomial distribution. By comparing the localized theoretical results with actual test results, a judgment standard for single-bottle abnormality alarms for each hospital is dynamically generated. This solves the technical bottleneck of traditional static threshold algorithms, which cannot adapt to differences in patient populations and different regions, reducing the false alarm rate for single-bottle abnormalities. Furthermore, real-time monitoring of single-bottle reagent quality improves the detection rate of single-bottle abnormalities and avoids unreliable test results.

[0044] Please refer to Figure 1 Some embodiments provide a method for dynamic monitoring of the quality of a single bottle of reagent, which includes the following steps:

[0045] Step S100: Obtain the detection result of the current sample after the current reagent performs the corresponding detection item on the current sample, and obtain the multi-source detection background data of the current sample.

[0046] Considering that in addition to patient information and testing season (or testing date) affecting the test results of the current reagent, the testing hospital (or the region corresponding to the testing hospital) where the current sample is located also has a certain impact. For example, if a certain disease is more common in a region, hospitals in that region often have better treatment results for that disease. Therefore, the positive rate of the test results obtained by hospitals in that region for that disease is usually higher than the positive rate of the corresponding test results obtained by hospitals in other regions. Therefore, in order to ensure that reagent quality monitoring can adapt to differences in patient population, region, season, etc., the multi-source testing background data to be obtained in this embodiment includes one or more of the following: patient characteristic data, testing hospital, and testing date. Among them, patient characteristic data includes one or more of the following: patient age group, gender, disease label, and patient region.

[0047] For example, in a hospital, taking tumor disease as an example, assuming that the current reagent is used to detect tumor markers, the current sample is tested for the corresponding tumor marker detection item using the current reagent, and the test result (positive or negative) of the current sample for that detection item is obtained; at the same time, the multi-source detection background information of the current sample is obtained through the hospital's LIS system (i.e., laboratory information management system), so as to obtain data such as the patient's age, gender, tumor disease label, test date, test hospital, and patient's region corresponding to the current sample.

[0048] The patient's age is determined by their age group, for example, 0 to 11 years old is the infant stage, 12 to 18 years old is the adolescent stage, 19 to 35 years old is the young adult stage, 36 to 59 years old is the middle-aged stage, and 60 years old and above is the elderly stage. The tumor disease label refers to the type of tumor the patient has, with each type of tumor corresponding to a disease label. The patient's region refers to the region of origin of the patient (such as the patient's place of residence), thereby combining factors such as the geographical environment of the patient's place of residence to improve the reliability of subsequent reagent quality monitoring.

[0049] It should be noted that since the core components of clinical testing reagents are mostly bioactive substances such as enzymes, antibodies, and primers / probes, they are extremely sensitive to temperature and can only function normally within a specific temperature range. Therefore, under abnormal conditions such as equipment malfunction (e.g., fully automated biochemical analyzers, immunoassay analyzers) or reagent tray temperature exceeding the specified requirements, even if the current reagent quality is normal, the corresponding test results will be biased and cannot reflect the actual quality of the reagent. Therefore, this embodiment requires ensuring that the equipment and reagent tray temperature are under normal conditions before using the current reagent to perform the corresponding test on the current sample and obtain the multi-source detection background information corresponding to the current sample.

[0050] Step S110: Based on the multi-source detection background data and detection results of the current sample, obtain the feature vector of the current sample.

[0051] In this embodiment, feature encoding is performed based on the multi-source detection background data and detection results of the current sample to convert discrete classification features into numerical features, thereby obtaining the feature vector of the current sample.

[0052] For example, the multi-source detection background data and corresponding detection results of the current sample can be feature-encoded using one-hot encoding, binary encoding, or multi-hot encoding. This converts the multi-source detection background data of the current sample into numerical values, allowing the multi-source detection background data to be fused with the detection results to obtain the feature vector corresponding to the current sample. Subsequently, the positive probability of the current sample under normal conditions can be predicted based on the obtained feature vector, thereby determining whether there is an abnormality in the quality of the current reagent.

[0053] This embodiment uses unique thermal encoding to encode features based on the patient's age group, the season of testing, and the patient's location. The feature encoding results for the patient in different age groups (e.g., infants, children, young adults, middle-aged, and elderly) can be represented as: [1, 0, 0, 0, 0], [0, 1, 0, 0, 0], [0, 0, 1, 0, 0], [0, 0, 0, 1, 0], [0, 0, 0, 0, 1]. The testing season is determined by the testing date. For example, January-March is spring, April-June is summer, July-September is autumn, and October-December is winter. The feature encoding results for different testing seasons can be represented as: [1, 0 ... [0,0], [0,1,0,0], [0,0,1,0], [0,0,0,1]; The patient's region is determined according to the patient's place of origin. In this embodiment, the patient's region is divided into seven major geographical regions: East China, South China, North China, Central China, Southwest China, Northwest China, and Northeast China. Then, the patient's region is feature-coded by one-hot coding. The feature coding results corresponding to different patient regions can be represented as follows: [1,0,0,0,0,0,0], [0,1,0,0,0,0,0], [0,0,1,0,0,0,0], [0,0,0,1,0,0,0], [0,0,0,1,0,0,0], [0,0,0,0,0,1,0], [0,0,0,0,0,1].

[0054] The current sample is characterized by binary coding, which encodes the patient's gender (male or female) and the test result (negative or positive). For example, when the patient is male, it is encoded as 0, and when the patient is female, it is encoded as 1. After the current reagent performs the corresponding test on the current sample, when the test result is negative, it is encoded as 0, and when the test result is positive, it is encoded as 1.

[0055] Since a patient may suffer from multiple diseases simultaneously, multi-hot coding can be used to characterize the patient's disease labels. Taking tumor disease labels as an example, according to the cancer classification in the ICD-11 document (a globally unified standard document for the classification and coding of diseases and health problems), there are eight types of tumor diseases, including liver cancer, pancreatic cancer, stomach cancer, colorectal cancer, lung cancer, ovarian cancer, breast cancer, and prostate cancer. Each type of tumor disease corresponds to a disease label. If a hospital has eight types of tumor disease labels, then if the patient in the current sample has the first type of tumor disease (i.e., liver cancer), the patient's disease label will be [1, 0, 0, 0, 0, 0, 0, 0]. If the patient in the current sample has the first and fourth types of tumor diseases (i.e., liver cancer and colorectal cancer), the patient's disease label will be [1, 0, 0, 1, 0, 0, 0, 0], and so on.

[0056] This embodiment uses feature encoding to integrate multi-source background data into a high-dimensional feature vector by combining the patient's age group, gender, disease label, testing season, region, and test results. If the test item corresponding to the current reagent is a tumor marker test item, the final feature vector of the current sample is a 26-dimensional feature vector. Subsequently, the positive probability of the current sample under normal reagent quality conditions is predicted based on this feature vector to monitor the reagent quality.

[0057] Step S120: Obtain the positive probability of the current sample based on the feature vector of the current sample.

[0058] In this embodiment, the feature vector of the current sample is input into the probability prediction model of the test item corresponding to the current reagent to obtain the positive probability of the current sample under the assumption that the quality of the current reagent is normal; wherein, the probability prediction model of the test item corresponding to the current reagent is trained by the following method, such as Figure 2 As shown, it includes the following steps:

[0059] Step S121: Obtain multi-source datasets and device log data.

[0060] This embodiment obtains a multi-source dataset from the historical medical records of the hospital where the current sample is being tested. It can also be the test data from multiple hospitals in the same region, such as the test data from 5 hospitals in the same province over the past 3 years. The resulting multi-source dataset contains the multi-source test background data for each sample, as well as the test results obtained after each sample has undergone the corresponding test using multiple reagents.

[0061] It should be noted that this embodiment further integrates the obtained multi-source datasets with the disease epidemic trends released by the CDC (such as the incidence rate of a certain disease in a specific region or time period). The disease epidemic trends can be obtained through the disease spectrum. When the positive rate of a certain reagent in the multi-source dataset is abnormal, the disease epidemic trend can be compared to quickly determine whether it is due to a problem with the quality of the reagent itself or a normal fluctuation caused by an increase in the actual prevalence of the disease. This ensures the data quality in the multi-source dataset and thus constructs the detection effect of each reagent in each detection project.

[0062] For example, if the increase in the positive rate of a test result from a vial of reagents matches the increase in the incidence rate of the disease published by the CDC, then it is considered that the increase in the positive rate of the reagent is caused by the increase in the actual prevalence of the disease, and the quality of the vial of reagents itself is normal. However, if the positive rate increases, but the actual incidence rate of the disease does not increase, then it is considered that the quality of the reagent itself is abnormal. In this case, the test results corresponding to the reagent in the multi-source dataset need to be removed so that they do not participate in the subsequent training process of the probability prediction model, in order to ensure the accuracy of the positive rate predicted by the probability prediction model.

[0063] In addition, it is necessary to obtain equipment log data, which is used to determine whether the equipment is under abnormal conditions, such as whether the equipment is malfunctioning or whether the reagent tray temperature exceeds the specified requirements. The equipment log is used to obtain the temperature change of the reagent tray and to statistically analyze the test results of each item and each bottle of reagent in the multi-source dataset under each temperature change. When the reagent tray temperature exceeds the specified requirements, for example, when the reagent tray temperature is displayed as 20 degrees, the equipment is considered to be under abnormal conditions.

[0064] Step S122: Preprocess the multi-source dataset based on device log data.

[0065] After obtaining the multi-source dataset, it is preprocessed based on the device log data. In this embodiment, the preprocessing of the multi-source dataset includes outlier removal and data association. Outlier removal is used to remove test results in the multi-source dataset that are under abnormal device conditions based on the device log data. Data association is used to associate the test result corresponding to each reagent with the multi-source detection background data of the corresponding sample, thereby establishing the correspondence between each test result obtained from each bottle of reagent and the multi-source detection background data of its corresponding sample.

[0066] For ease of description, this embodiment will still refer to the preprocessed multi-source dataset as the multi-source dataset. Subsequently, based on the preprocessed multi-source dataset, the detection result corresponding to each sample in the detection item will be extracted to train the probability prediction model corresponding to each detection item.

[0067] Step S123: Obtain the probability prediction model corresponding to each detection item based on the preprocessed multi-source dataset.

[0068] This embodiment trains a model based on the multi-source detection data and results corresponding to each detection item in the obtained multi-source dataset to obtain the probability detection model corresponding to each detection item, realizing localized real-time training in hospitals, so as to obtain the positive probability of each sample under the conditions of different patient groups, different seasons, and different regions, assuming that the reagent quality is normal.

[0069] For any given test item, taking the current reagent's corresponding test item (tumor marker test item) as an example, the process is as follows: Extract the test results for each sample corresponding to this test item from the multi-source dataset, thus extracting all test results for the current reagent's corresponding test item. All the obtained test results can be stored in order of reagent number and test date. The multi-source dataset can be divided into training and testing sets according to a certain ratio. For example, 90% of the data in the multi-source dataset can be used as the training set, and 10% as the testing set. The model is trained using the training set, and its generalization ability is evaluated using the testing set, ultimately yielding the probabilistic prediction model corresponding to this test item.

[0070] Then, based on the detection results of each sample in the detection project and the multi-source detection background data, the feature vector of each sample is obtained. For example, the multi-source detection background data and the corresponding detection results of each sample are feature-encoded by one-hot encoding, binary encoding or multi-hot encoding, so as to convert the multi-source detection background data of each sample into specific values, so as to fuse the multi-source detection background data and the detection results to obtain the feature vector corresponding to each sample. The feature encoding method in this embodiment has been specifically explained in step S110, and will not be repeated here.

[0071] The feature vector of each sample is input into a pre-defined network to obtain the positive probability of that sample. For example, the pre-defined neural network is an MLP (Multilayer Perceptron) neural network, including one input layer, two hidden layers and one output layer. The first hidden layer is configured with 64 neurons (nodes) and the second hidden layer is configured with 32 neurons (nodes). The activation function of the two hidden layers is ReLU, and the Dropout regularization mechanism is introduced. The number of neurons (nodes) determines the ability of the layer to represent the input features. The ReLU activation function is used to introduce non-linear characteristics into the model, allowing the model to learn complex data patterns (such as non-linear correlations in reagent testing data). Dropout prevents the model from over-relying on specific features by randomly disabling some nodes during training, thereby preventing overfitting and improving the model's generalization ability on new data.

[0072] It should be noted that the prediction of the positive probability can also be achieved using neural networks such as Lightweight Gradient Boosting Machine (LightGBM), Extreme Gradient Boosting Machine (XGBoost), Random Forest, and Linear Regression.

[0073] The loss function is calculated based on the positive probability and test result of each sample. For example, the binary cross-entropy (BCE) is used as the loss function. The test result obtained by each sample through reagent testing is used as the true label of the sample. The binary cross-entropy is calculated based on the positive probability and true label of each sample. The preset network is trained based on the obtained loss function (binary cross-entropy). The convergence of the obtained loss function determines whether to continue training or stop training the preset model. When the loss function converges, the preset model is considered to have completed training, and the probability prediction model of the test item corresponding to the current reagent is obtained.

[0074] Step S130: Obtain historical samples before the current sample, and obtain the positive probability and test results of the historical samples. Calculate the abnormal probability of the current reagent based on the positive probability and test results of the current sample and the historical samples.

[0075] Obtain a preset number (e.g., N-1) of historical samples prior to the current sample. These historical samples include those tested with the current reagent and / or reagents of the same type as the current reagent before the current sample. For example, if the number of historical samples tested with the current reagent before the current sample is less than N-1, the historical samples corresponding to the current sample can be determined based on the testing date of the current sample and the samples corresponding to other reagents of the same type as the current reagent (the previous bottle of reagent). At this point, the historical samples corresponding to the current sample have already obtained test results, and the positive probability corresponding to the sample has been obtained through the probability prediction model corresponding to the current test item. That is, the historical samples corresponding to the current sample also each have a positive probability and a test result.

[0076] Considering that the probability of a positive test result for different patients is affected by various factors such as their own condition and the season, meaning that the probability of a positive test result for different patients is not the same, this embodiment uses the Poisson-Binomial distribution to determine whether there is an abnormality in the quality of the current reagent. The Poisson-Binomial distribution refers to the probability distribution of the total number of successes in a series of independent Bernoulli tests when the success probabilities of these tests are different.

[0077] In this embodiment, when the total number of positive test results is K in the current sample and historical samples (a total of N samples), the distribution is used to determine the normal situation. The probability (i.e., the significance probability) that the number of positive results in the test results corresponding to the current reagent is at least K is calculated. By comparing the obtained significance probability with the preset significance level, it is determined whether the test result of the current reagent is a test result under normal circumstances, thereby enabling real-time monitoring of reagent quality.

[0078] In this embodiment, the number of positive samples is determined based on the test results of the current sample and historical samples, and denoted as K. Then, the significance probability (P-value) is calculated based on the number of positive samples K and the positive probabilities corresponding to the current sample and historical samples. The obtained significance probability is used as the anomaly probability of the current reagent, which characterizes the degree of consistency between the actual test results of the current reagent and the corresponding test results under the assumption that the quality of the current reagent is normal. The anomaly probability of the current reagent can be expressed as:

[0079]

[0080] Where P{≥K positive results} represents the probability that the number of positive results in the current reagent's test results is at least K, i.e., the significance probability (or anomaly probability) of the current reagent; A represents the subset of the current sample and its historical samples, p i p represents the positive probability of the i-th sample in subset A; jLet represent the positive probability of the j-th sample among the remaining samples excluding subset A.

[0081] Step S140: Determine whether there is an abnormality in the quality of the current reagent based on the abnormality probability of the current reagent.

[0082] The quality of the current reagent is monitored by determining whether there is an abnormality based on the current reagent's anomaly probability and a preset significance level. If the anomaly probability of the current reagent is less than the preset significance level, it indicates that the contradiction between the actual test result of the current reagent and the test result under the assumed normal condition is more significant, and the null hypothesis is more likely to be invalid, meaning that the quality of the current reagent is more likely to be abnormal. Conversely, if the anomaly probability of the current reagent is greater than the test result under the assumed normal condition, it indicates that the consistency between the actual test result of the current reagent and the test result under the assumed normal condition is higher, and the null hypothesis is more likely to be valid, meaning that the quality of the current reagent is more likely to be normal.

[0083] For example, the preset significance level α can be set to 0.001. If the abnormal probability corresponding to the current reagent is less than 0.001, the current reagent is considered to be abnormal, and a single-bottle reagent abnormality alarm is triggered. Otherwise, the test result of the current reagent for the next sample is obtained, and the corresponding positive probability is obtained based on the test result of the next sample and the multi-source detection background data, so as to continue to monitor the quality of the current reagent.

[0084] For example, assuming N=10, meaning there are 10 samples in total (including the current sample and its historical samples), with positive probabilities of [0.15, 0.08, 0.10, 0.20, 0.05, 0.07, 0.12, 0.09, 0.18, 0.06], and 8 of these samples have positive results (K=8), then the probability of at least K positive results based on the positive probabilities of the current sample and its historical samples is 2.1 × 10⁻⁶. -6 If the value is less than the preset significance level α, it is determined that the current reagent is abnormal and an abnormal alarm is triggered.

[0085] It should be noted that, considering the large amount of data related to disease types and corresponding testing items, this embodiment only uses tumor disease types and tumor marker testing items as examples for illustration. However, other disease types in the ICD-11 document (a globally unified standard document for classifying and coding diseases and health problems), such as diabetes and endocrine diseases, can also be monitored for abnormalities in the quality of single-bottle reagents using this method.

[0086] This embodiment obtains the test results of each sample after testing with a single bottle of reagent, as well as the corresponding multi-source detection background data for that sample. Based on the hospital's localized probability detection model, it obtains the positive probability of each sample under normal conditions, considering factors such as the patient's own characteristics, region, and testing season. Then, based on the positive probabilities of the current sample and historical samples and the test results, it calculates the significance probability under the assumption that the reagent quality is normal in real time, thereby realizing the real-time adjustment of the dynamic P-value (specificity threshold) of a single bottle of reagent and thus monitoring the current reagent quality in real time. By fusing test results with multi-source background data, the assessment of reagent quality becomes more realistic, overcoming the technical bottleneck of traditional static threshold methods that cannot adapt to differences in patient populations, regions, and seasons, and reducing the false alarm rate of single-bottle abnormalities. By predicting the positive probability of the current sample under normal reagent quality after each test result obtained with the current reagent, and combining the positive probability of historical samples with test results, hospital-specific thresholds are dynamically generated, thereby automatically monitoring the quality of the current reagent in real time without manual intervention. This avoids unreliable test results caused by the lag in adjusting reagent quality monitoring thresholds in traditional methods, while still ensuring quality control for each bottle of reagent.

[0087] Some embodiments of the present invention also disclose a computer-readable storage medium comprising a program that can be executed by a processor to implement the methods described in any of the embodiments herein.

[0088] Those skilled in the art will understand that all or part of the functions of the various methods in the above embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, which may include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to achieve the above functions. For example, the program can be stored in the memory of a device, and when the program in the memory is executed by the processor, all or part of the above functions can be achieved. In addition, when all or part of the functions in the above embodiments are implemented by computer programs, the program can also be stored in a server, another computer, disk, optical disk, flash drive, or external hard drive, etc., and can be downloaded or copied to the memory of a local device, or the system of the local device can be updated. When the program in the memory is executed by the processor, all or part of the functions in the above embodiments can be achieved.

[0089] The above examples illustrate the present invention only to aid in understanding it and are not intended to limit the scope of the invention. Those skilled in the art can make various simple deductions, modifications, or substitutions based on the principles of this invention.

Claims

1. A method for dynamic monitoring of the quality of a single bottle of reagent, characterized in that, include: Obtain the test results for the current sample after performing the corresponding test on the current sample with the current reagent; Obtain multi-source detection background data for the current sample, wherein the multi-source detection background data includes one or more of patient characteristic data, testing hospital, and testing date; Based on the multi-source detection background data and detection results of the current sample, the feature vector of the current sample is obtained; The positive probability of the current sample is obtained based on the feature vector of the current sample; Obtain historical samples preceding the current sample, and obtain the positive probability and detection results of the historical samples; The abnormal probability of the current reagent is calculated based on the positive probability and test results of the current sample and the historical samples. Determine whether there is an abnormality in the quality of the current reagent based on the probability of abnormality of the current reagent.

2. The method for dynamic monitoring of single-bottle reagent quality as described in claim 1, characterized in that, The patient characteristic data includes one or more of the patient's age group, gender, disease label, and patient's region.

3. The method for dynamic monitoring of single-bottle reagent quality as described in claim 2, characterized in that, The step of obtaining the feature vector of the current sample based on the multi-source detection background data and detection results includes: Based on the multi-source detection background data and detection results of the current sample, feature encoding is performed to convert discrete classification features into numerical features, thereby obtaining the feature vector of the current sample.

4. The method for dynamic monitoring of single-bottle reagent quality as described in claim 1, characterized in that, The step of obtaining the positive probability of the current sample based on the feature vector of the current sample includes: The feature vector of the current sample is input into the probability prediction model of the test item corresponding to the current reagent to obtain the positive probability of the current sample under the assumption that the quality of the current reagent is normal. The probability prediction model for the current reagent's corresponding test item is trained using the following methods: Obtain a multi-source dataset, which contains multi-source detection background data for each sample and the detection results obtained after each sample has been tested for corresponding detection items using multiple reagents; For the current reagent-corresponding test item: extract the test result corresponding to each sample for the test item based on the multi-source dataset; obtain the feature vector of each sample based on the test result corresponding to each sample for the test item and the multi-source test background data; input the feature vector of each sample into a preset network to obtain the positive probability corresponding to the sample; calculate the loss function based on the positive probability corresponding to each sample and the test result; train the preset network based on the loss function until the loss function converges to obtain the probability prediction model corresponding to the test item.

5. The method for dynamic monitoring of single-bottle reagent quality as described in claim 4, characterized in that, Also includes: Obtain device log data, which is used to determine whether the device is under abnormal conditions.

6. The method for dynamic monitoring of single-bottle reagent quality as described in claim 5, characterized in that, After acquiring the multi-source dataset, and before extracting the detection result corresponding to each sample for the detection item based on the multi-source dataset, the process also includes: The multi-source dataset is preprocessed, including outlier removal and data association. The outlier removal is used to remove test results in the multi-source dataset that are under abnormal equipment conditions based on equipment log data; the data association is used to associate the test results corresponding to each reagent with the multi-source test background data of the corresponding sample.

7. The method for dynamic monitoring of single-bottle reagent quality as described in claim 1, characterized in that, The acquisition of historical samples prior to the current sample includes: acquiring a preset number of historical samples prior to the current sample, wherein the historical samples include samples that were tested prior to the current sample by the current reagent and / or reagents of the same type as the current reagent.

8. The method for dynamic monitoring of single-bottle reagent quality as described in claim 7, characterized in that, The calculation of the abnormal probability of the current reagent based on the positive probability and test results of the current sample and the historical samples includes: The number of positive samples is determined based on the test results corresponding to the current samples and the historical samples. The significance probability is calculated based on the number of positive samples and the positive probabilities corresponding to the current sample and the historical samples; the significance probability is used as the anomaly probability of the current reagent.

9. The method for dynamic monitoring of single-bottle reagent quality as described in claim 1, characterized in that, The step of determining whether the quality of the current reagent is abnormal based on the probability of abnormality of the current reagent includes: The current reagent quality is assessed based on its anomaly probability and a preset significance level to determine whether there is an anomaly. If an anomaly is detected, a single-bottle reagent anomaly alarm will be triggered; otherwise, the test results for the next sample using the current reagent will continue to be obtained.

10. A storage medium, characterized in that, The storage medium stores a computer program that can be executed by a processor to implement the method as described in any one of claims 1-9.

Images