Patient record matching method, storage medium, and device based on patient master index

Abstract

The present invention provides a patient record matching method based on a patient primary index, a storage medium and a device, the patient record matching method based on a patient primary index comprises: determining a training sample and a test sample based on a patient information set extracted from a patient information database; performing variable analysis on the attributes of the training sample respectively; comparing the training sample after the variable analysis with the test sample, and adjusting the similarity between the training sample and the test sample by the preference weight of the training sample; training a patient matching model by using the training sample to obtain the weight and threshold of the training sample; and evaluating the effect of the weight and the threshold by the test sample. The present invention saves manual workload and realizes a more scientific patient record matching process by automatically generating weights and thresholds with excellent effects.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical data processing, and relates to a patient record matching method, and in particular to a patient record matching method based on a patient master index, a storage medium and a device. Background Art

[0002] The Enterprise Master Patient Index (EMPI) is a patient basic information retrieval directory. Its purpose is to effectively associate multiple records in a system or multiple records in multiple systems. The core function of the Enterprise Master Patient Index is a matching algorithm based on the similarity of patient records. The similarity between a newly registered patient record and a patient record with a known EMPI number in the database is calculated. There are currently two matching algorithms in the prior art: one is a large weight priority algorithm, and the other is a comprehensive weight algorithm. Both algorithms require setting the weights of each patient attribute comparison result and matching upper and lower thresholds. Then, the two algorithms calculate the similarity of the two patient records through the comparison results of each attribute and the corresponding attribute comparison result weights. If the similarity exceeds the upper threshold, they are merged; if the similarity is lower than the lower threshold, a new patient is registered. In both algorithms, manual judgment is more involved.

[0003] The defects of the above existing solutions mainly include the following aspects: (1) The weight of the attribute comparison result is set manually. (2) The upper and lower thresholds are set manually. The manual setting of weights and thresholds relies on subjective experience and lacks scientific basis. Moreover, it is necessary to iterate and adjust the weights and thresholds multiple times to achieve the best effect of the EMPI patient matching algorithm, which requires a huge amount of manual work.

[0004] Therefore, how to provide a patient record matching method, storage medium and device based on the patient master index to solve the defects of the existing technology that it is unable to save manual workload in the patient record matching process and achieve more scientific matching has become a technical problem that needs to be urgently solved by technical personnel in this field. Summary of the invention

[0005] In view of the shortcomings of the prior art described above, the purpose of the present invention is to provide a patient record matching method, storage medium and device based on patient master index, so as to solve the problem that the prior art cannot save manual workload in the patient record matching process and achieve more scientific matching.

[0006] To achieve the above-mentioned purpose and other related purposes, the present invention provides a patient record matching method based on patient primary index on one hand, characterized in that the patient record matching method based on patient primary index includes: determining training samples and test samples based on patient information sets extracted from a patient information database; performing variable analysis on the attributes of the training samples respectively; comparing the training samples after variable analysis with the test samples, and adjusting the similarity between the training samples and the test samples through the preference weights of the training samples; training a patient matching model using the training samples to obtain weights and thresholds of the training samples; and evaluating the effects of the weights and thresholds through the test samples.

[0007] In one embodiment of the present invention, the step of determining training samples and test samples based on a patient information set extracted from a patient information database includes: extracting the training samples using preset extraction conditions, and labeling the training samples with unique identification information; extracting the test samples using matching candidate rules of an actual hospital, and labeling the test samples with the unique identification information.

[0008] In one embodiment of the present invention, the step of performing variable analysis on the attributes of the training samples respectively includes: performing univariate analysis on the attributes of the training samples, the univariate analysis content includes: missing rate and variance; performing multivariate analysis on the attributes of the training samples, the multivariate analysis content includes: Pearson correlation coefficient and variance inflation factor; eliminating unfavorable attributes of the training samples through the results of the univariate analysis and the multivariate analysis; in response to the attributes of the training samples being multivariate variables, performing dimensionality reduction processing on the multivariate variables.

[0009] In one embodiment of the present invention, the step of comparing the training samples after variable analysis and the test samples, and adjusting the similarity between the training samples and the test samples by the preference weights of the training samples, includes: calculating the density ratio of the training samples to the test samples by transfer learning; the number of the test samples is the same as that of the training samples, and the density ratio is the preference weights of the training samples during the training of the patient matching model.

[0010] In one embodiment of the present invention, the step of using the training samples to train the patient matching model and obtaining the weights and thresholds of the training samples includes: using the training samples to obtain the patient matching model based on a supervised learning model; standardizing the total score of the weight of the patient matching model; calculating the similarity of each training sample from 0 to 100 points; determining candidate threshold pairs using primary statistical indicators, and determining upper and lower thresholds based on secondary statistical indicators.

[0011] In one embodiment of the present invention, the step of evaluating the effectiveness of the weights and the thresholds through the test samples includes: obtaining user-defined weights and custom thresholds; generating custom statistical indicators of the custom weights and the custom thresholds on the test samples; comparing the statistical indicators of the patient matching model with the custom statistical indicators; the statistical indicators include: accuracy, percentage of manual participation, recall rate, false positive rate and F1 value.

[0012] In one embodiment of the present invention, the patient record matching method based on the patient primary index also includes: visually presenting the results of the effect evaluation, and the visual presentation method includes: at least one of text, pictures, tables or statistical charts.

[0013] In one embodiment of the present invention, the step of visually presenting the results of the effect evaluation includes: performing visual web access to the statistical indicator results of the weight and the threshold in the test sample.

[0014] To achieve the above-mentioned purpose and other related purposes, the present invention provides, on the other hand, a computer-readable storage medium having a computer program stored thereon, which implements the patient record matching method based on the patient primary index when executed by a processor.

[0015] To achieve the above-mentioned purpose and other related purposes, the last aspect of the present invention provides an electronic device, including: a processor and a memory; the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the electronic device executes the patient record matching method based on the patient primary index.

[0016] As described above, the patient record matching method, storage medium and device based on the patient master index of the present invention have the following beneficial effects:

[0017] The present invention performs variable analysis processing on the extracted training samples, and uses preference weights to make the processed training samples closer to the actual test samples, thereby automatically generating weights and thresholds with excellent effects, saving manual workload, and realizing a more scientific patient record matching process. The present invention does not need to spend a lot of time on iterative adjustment of weights and thresholds. The training samples and test samples are automatically generated and annotated by the present invention, and only a small number of statistical indicators are set. The weights and thresholds are automatically obtained through supervised learning, which reduces a lot of manual workload. And through statistical indicators, weights and thresholds are automatically selected. The statistical results of the test samples based on the matching algorithm of this weight and threshold are further visualized, which can be compared with matching algorithms that select other weights and thresholds to understand the advantages and disadvantages of various weight threshold selections, so as to facilitate confirmation of the rationality of various selections. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Shown is a principle flow chart of a patient record matching method based on a patient master index in one embodiment of the present invention.

[0019] Figure 2 FIG. 1 is a schematic diagram showing data processing in an embodiment of a patient record matching method based on a patient primary index according to the present invention.

[0020] Figure 3A Shown is a score distribution diagram of a random sampling matching algorithm in an embodiment of a patient record matching method based on a patient primary index of the present invention.

[0021] Figure 3B Shown is a score distribution diagram of a conditional sampling matching algorithm in an embodiment of a patient record matching method based on a patient primary index of the present invention.

[0022] Figure 4 Shown is a distribution diagram of the comparison results of telephone numbers in training samples and actual test samples in an embodiment of the patient record matching method based on patient primary index of the present invention.

[0023] Figure 5 It is a schematic diagram showing the structural connection of an electronic device in one embodiment of the present invention.

[0024] Component number description

[0025] 5 Electronic devices

[0026] 51 Processor

[0027] 52 Memory

[0028] Steps S11 to S15 DETAILED DESCRIPTION

[0029] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0030] It should be noted that the illustrations provided in the following embodiments are only used to illustrate the basic concept of the present invention in a schematic manner, and thus the illustrations only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0031] The patient record matching method, storage medium and device based on the patient primary index of the present invention save manual workload and realize a more scientific patient record matching process by automatically generating weights and thresholds with excellent effects.

[0032] The following will be combined Figures 1 to 5 The principles and implementation methods of a patient record matching method, storage medium and device based on a patient primary index of this embodiment are explained in detail so that those skilled in the art can understand the patient record matching method, storage medium and device based on a patient primary index of this embodiment without creative work.

[0033] See also Figure 1 , which is a principle flow chart of a patient record matching method based on a patient master index in one embodiment of the present invention. Figure 1 As shown, the patient record matching method based on the patient primary index specifically includes the following steps:

[0034] S11, determining training samples and test samples based on a patient information set extracted from a patient information database.

[0035] In one embodiment, S11 specifically includes the following steps:

[0036] (1) Extracting the training samples using preset extraction conditions, and marking the training samples using unique identification information.

[0037] Specifically, see Figure 2 , which is a schematic diagram of data processing in an embodiment of the patient record matching method based on the patient master index of the present invention. Figure 2 As shown in the figure, the patient information set is automatically extracted from the patient information table in the hospital database to generate training samples, and an appropriate amount of training samples are extracted through different extraction conditions, and then automatically labeled with unique identification information. For example, the names are the same, or the names and genders are the same, or the genders and dates of birth are the same, that is, different attribute combinations are generated according to different extraction conditions.

[0038] In practical applications, in order to reduce manual workload, the extracted samples must include a unique identifier for automatic labeling. 1 indicates the same identifier, and 0 indicates different identifiers. That is, two pieces of data with the same identifier will be compared, and the comparison result and the label 1 will generate a training sample. In the embodiment, the unique identification information such as the ID card number or outpatient number, hospitalization number, medical insurance number, etc. is used as the unique identifier as the labeling standard. Secondly, please refer to Figure 3A , which shows the score distribution diagram of the random sampling matching algorithm in one embodiment of the patient record matching method based on the patient master index of the present invention. Figure 3AAs shown in the figure, the horizontal axis is the similarity score and the vertical axis is the number of samples. Assuming there are N patient records, if they are matched two by two, a Cartesian product N will be formed. 2 The dataset is too large, which greatly increases the training time. In addition, the proportion of 0 samples in the sample is much greater than that of 1 samples, resulting in an extremely unbalanced sample ratio. The model trained with this sample will extremely predict the result as 0 samples. Even if the sample ratio is balanced, the similarity score will be extremely polarized. Please refer to Figure 3B , which shows the score distribution diagram of the conditional sampling matching algorithm in one embodiment of the patient record matching method based on the patient master index of the present invention. Figure 3B As shown in the figure, the horizontal axis is the similarity score and the vertical axis is the number of samples. According to the conditions (whether the names are the same, whether the genders are the same, whether the birth dates are the same, etc.), one sample of 0 and one sample of 1 that meet the conditions are extracted at a time. The final result of the sample is that the ratio of 0 and 1 samples is 1:1, and the similarity scores are evenly distributed between 0-100.

[0039] (2) The test sample is extracted according to the matching candidate rule of the actual hospital, and the test sample is labeled according to the unique identification information.

[0040] Specifically, the automatic extraction of actual test samples is implemented by SQL (Structured Query Language) and Python. An appropriate amount of actual test samples and their annotations are extracted through the actual hospital's matching candidate rules. Among them, due to the huge amount of personal information records in the hospital database, it is impossible to match them one by one, so the hospital will customize the matching candidate rules. For example, the matching candidate rules of a hospital are the same name and gender, so as to generate a candidate queue, reduce the matching amount, and increase the matching efficiency.

[0041] Therefore, step S11 solves the problem of requiring a large amount of human resources for manual labeling in order to obtain the training samples and test samples required for the artificial intelligence model.

[0042] S12, performing variable analysis on the attributes of the training samples respectively.

[0043] In one embodiment, S12 specifically includes the following steps:

[0044] (1) Performing univariate analysis on the attributes of the training samples, wherein the univariate analysis includes missing rate and variance.

[0045] Specifically, the missing rate of each attribute in the training sample is obtained, and automatically screened by a preset missing threshold. For example, for each attribute (name, gender, date of birth, etc.), the attributes exceeding the missing rate threshold are discarded by a preset missing rate threshold.

[0046] Specifically, the variance of each attribute in the training sample is obtained, and the influence of the corresponding attribute on the prediction result of the patient matching model is judged according to the size of the variance. For example, the variance of the remaining attributes in the sample after the missing rate screening is screened, and the attributes with too small variance are filtered out, which means that the attribute values ​​are almost all the same.

[0047] It should be noted that the missing rate and variance are merely examples of implementation methods for a single variable, and other indicators of single variable analysis are also included in the scope of protection of the present invention.

[0048] (2) Performing a multivariate analysis on the attributes of the training samples, wherein the multivariate analysis includes: Pearson correlation coefficient and variance inflation factor.

[0049] Specifically, the Pearson coefficient or chi-square test is performed on each pair to screen out the attributes with too high correlation coefficients. Finally, the remaining attributes after screening need to calculate the variance expansion factor to screen out the attributes with too high variance inflation coefficient or multicollinearity.

[0050] It should be noted that the Pearson correlation coefficient and the variance inflation factor are merely exemplary implementations for multivariate analysis, and other multivariate analysis indicators are also included in the scope of protection of the present invention.

[0051] (3) Eliminating unfavorable attributes of the training samples through the results of the univariate analysis and the results of the multivariate analysis.

[0052] Specifically, for the missing rate, the attribute missing rate is too high, that is, there are too few valid data. For the variance, the attribute variance is too small, that is, this attribute value can be regarded as a constant and does not affect the model prediction results. For the Pearson correlation coefficient and variance inflation coefficient, the Pearson correlation coefficient is too high and the variance inflation coefficient is too high, that is, there is collinearity between the attributes. Collinearity will cause spatial instability of the solution, resulting in weak generalization ability of the model. In practical applications, by reasonably setting the thresholds of missing rate, variance, Pearson correlation coefficient and variance inflation coefficient, unfavorable attributes are determined and then eliminated.

[0053] (4) In response to the attribute of the training sample being a multivariate variable, performing dimensionality reduction processing on the multivariate variable.

[0054] Specifically, through dimensionality reduction processing, the amount of calculation during the training of the patient matching model can be reduced, and the training time can be shortened, making the model effect more excellent.

[0055] S13, comparing the training sample after the variable analysis with the test sample, and adjusting the similarity between the training sample and the test sample according to the preference weight of the training sample.

[0056] In one embodiment, transfer learning is used to calculate the density ratio of the training samples to the test samples; the number of the test samples is the same as the number of the training samples, and the density ratio is the preference weight of the training samples during the training of the patient matching model.

[0057] Specifically, since the distribution of the attribute comparison results of the training samples and the test samples are different, the model's good performance on the training samples does not mean that it will perform well on the test samples. Figure 4 , which shows the distribution diagram of the comparison results of telephone numbers in the training sample and the actual test sample in the patient record matching method based on the patient primary index of the present invention. Figure 4 As shown, since the attributes of the training sample and the test sample are the same, the attribute comparison distribution is different, which means that the distribution of the training sample and the test sample is different, which will cause the model that performs well on the training sample to perform poorly on the actual test sample. In an embodiment, sample-based non-inductive transfer learning will be used to calculate the preference weight of each training sample. The more similar the training sample is to the actual test sample, the higher the preference weight. Thus, the influence of each training sample in the training model is adjusted by the preference weight, so that the patient matching model obtained by training works better on the test sample. In practical applications, the process of training the patient matching model is actually to minimize the sum of the loss functions of all samples. By default, the preference weights are all 1, that is, the arithmetic mean. In the case of preference weights, the loss function of each sample will be multiplied by the preference weight and summed, that is, the weighted average.

[0058] S14, using the training samples to train a patient matching model, and obtaining weights and thresholds of the training samples.

[0059] In one embodiment, S14 specifically includes the following steps:

[0060] (1) Using the training samples, the patient matching model is obtained based on supervised learning model training.

[0061] Specifically, the patient matching model is obtained by training with training samples with preference weights based on a supervised learning model, and weights and thresholds are automatically obtained.

[0062] In practical applications, logistic regression in supervised learning is used to automatically obtain weights based on training samples with preference weights, where the weights are the weights of each attribute.

[0063] (2) The weights of the patient matching model are normalized to a total score, for example, the total score is normalized to 100.

[0064] (3) Calculate the similarity between 0 and 100 for each training sample.

[0065] (4) Determine candidate threshold pairs using primary statistical indicators and determine upper and lower thresholds based on secondary statistical indicators.

[0066] Specifically, the upper and lower thresholds are obtained by pre-set accuracy and percentage of manual participation, or other statistical indicators that can be used as substitutes, such as recall rate, false positive rate, F1 value, etc. In practical applications, the statistical indicator accuracy and percentage of manual participation of each pair of thresholds ((0, 1), (0, 2), (0, 3) ..... (98, 99), (98, 100), (99, 100)) are calculated, and the candidate threshold pairs that meet the pre-set accuracy and pre-set percentage of manual participation are first selected, and then the remaining statistical indicators are sorted to select the best pair. At this point, the present invention automatically obtains the comparison attributes required by the EMPI patient matching algorithm, the attribute comparison result weights, and automatically matches the upper and lower thresholds.

[0067] In actual applications, the accuracy is preset to 0.95 and the manual participation percentage is preset to 0.1. First, the candidate threshold pairs with a higher accuracy than the preset 0.95 and lower than the preset manual participation percentage 0.1 are selected, and then the remaining statistical indicators such as recall rate, false positive rate, and F1 value are sorted. The best pair is selected based on the higher recall rate, lower false positive rate, and higher F1.

[0068] S15, evaluating the effects of the weight and the threshold through the test sample.

[0069] In one embodiment, S15 specifically includes the following steps:

[0070] (1) Obtain user-defined weights and custom thresholds.

[0071] Specifically, a setting table or setting interface is provided to the user, for example, in response to the user filling in a custom weight and threshold in a table, the user-defined weight and custom threshold are obtained.

[0072] (2) Generate a custom statistical indicator based on the custom weight and the custom threshold on the test sample.

[0073] (3) Compare the statistical indicators of the patient matching model with the custom statistical indicators; the statistical indicators include: accuracy, percentage of manual participation, recall rate, false positive rate and F1 value. Specifically, among the statistical indicators of the actual test samples, the higher the accuracy is, the better, the lower the percentage of manual participation is, the better, the higher the recall rate is, the better, the lower the false positive rate is, the better, and the higher the F1 is, the better. The test results of the patient matching model for the test samples should be better than the custom results.

[0074] In one embodiment, after step S15, the patient record matching method based on the patient primary index further includes: visually presenting the results of the effect evaluation, and the visual presentation method includes: at least one of text, pictures, tables or statistical charts.

[0075] Furthermore, a visualized web access is performed for the statistical indicator results of the weight and the threshold in the test sample.

[0076] In practical applications, dash and flask are used to visualize the statistical indicator results of weights and thresholds in actual test samples for web access. Use tables, ROC (Receiver Operating Characteristic) curves, bar charts, etc. to visualize the results. After visualization, you can obtain the statistical data of the patient matching model on the training samples and actual test samples, as well as an input box for entering custom weights and thresholds. After confirming the custom weights and thresholds, statistical indicators of the custom weights and thresholds on the test samples will be generated. It can be further compared with patient matching models that select other weights and thresholds to understand the advantages and disadvantages of various weight threshold selections, so as to confirm the rationality of various selections.

[0077] The protection scope of the patient record matching method based on the patient primary index described in the present invention is not limited to the execution order of the steps listed in this embodiment. All solutions implemented by adding, reducing, or replacing steps in the prior art based on the principles of the present invention are included in the protection scope of the present invention.

[0078] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the patient record matching method based on the patient primary index is implemented.

[0079] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to the computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned computer-readable storage medium includes: ROM, RAM, magnetic disk or optical disk, etc., various computer storage media that can store program codes.

[0080] See also Figure 5 , which is a schematic diagram showing the structural connection of an electronic device in one embodiment of the present invention. Figure 5As shown, this embodiment provides an electronic device 5, which specifically includes: a processor 51 and a memory 52; the memory 52 is used to store computer programs, and the processor 51 is used to execute the computer programs stored in the memory 52, so that the electronic device 5 performs each step of the patient record matching method based on the patient primary index.

[0081] The above-mentioned processor 51 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0082] The memory 52 may include a random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0083] In practical applications, the electronic device may be a computer including all or part of the components such as a memory, a storage controller, one or more processing units (CPU), a peripheral interface, an RF circuit, an audio circuit, a speaker, a microphone, an input / output (I / O) subsystem, a display screen, other output or control devices, and an external port; the computer includes but is not limited to personal computers such as desktop computers, laptop computers, tablet computers, smart phones, and personal digital assistants (PDAs). In other embodiments, the electronic device may also be a server, which may be arranged on one or more physical servers according to various factors such as function and load, or may be a cloud server composed of a distributed or centralized server cluster, which is not limited in this embodiment.

[0084] In summary, the patient record matching method, storage medium and device based on the patient master index of the present invention perform variable analysis processing on the extracted training samples, and use preference weights to make the processed training samples closer to the actual test samples, thereby automatically generating weights and thresholds with excellent effects, saving manual workload, and realizing a more scientific patient record matching process. The present invention does not need to spend a lot of time on iterative adjustment of weights and thresholds. The training samples and test samples are automatically generated and annotated by the present invention, and only a small number of statistical indicators are set. The weights and thresholds are automatically obtained through supervised learning, which reduces a lot of manual workload. And the weights and thresholds are automatically selected through statistical indicators. The statistical results of the test samples based on the matching algorithm of this weight and threshold are further visualized, which can be compared with the matching algorithms that select other weights and thresholds to understand the advantages and disadvantages of various weight threshold selections, so as to facilitate the confirmation of the rationality of various selections. The present invention effectively overcomes the various shortcomings in the prior art and has a high industrial utilization value.

[0085] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by a person of ordinary skill in the art without departing from the spirit and technical concept disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A patient record matching method based on a patient primary index, characterized in that: The patient record matching method based on the patient primary index includes: Determine training samples and test samples based on a patient information set extracted from a patient information database; including: extracting the training samples using preset extraction conditions, and marking the training samples with unique identification information; extracting the test samples using matching candidate rules of actual hospitals, and marking the test samples with the unique identification information; Performing variable analysis on the attributes of the training samples respectively; Comparing the training sample after the variable analysis with the test sample, and adjusting the similarity between the training sample and the test sample by the preference weight of the training sample; The patient matching model is trained using the training samples to obtain the weights and thresholds of the training samples; including: using the training samples to obtain the patient matching model based on a supervised learning model; normalizing the total score of the weight of the patient matching model; calculating the similarity of each training sample from 0 to 100 points; determining candidate threshold pairs using primary statistical indicators, and determining upper and lower thresholds based on secondary statistical indicators; The effects of the weight and the threshold are evaluated through the test sample.

2. The patient record matching method based on the patient primary index according to claim 1, characterized in that: The step of performing variable analysis on the attributes of the training samples respectively comprises: Performing univariate analysis on the attributes of the training samples, wherein the univariate analysis includes: missing rate and variance; Performing a multivariate analysis on the attributes of the training samples, wherein the multivariate analysis includes: a Pearson correlation coefficient and a variance inflation factor; Eliminate unfavorable attributes of the training samples through the results of the univariate analysis and the results of the multivariate analysis; In response to the attribute of the training sample being a multivariate variable, a dimensionality reduction process is performed on the multivariate variable.

3. The patient record matching method based on patient primary index according to claim 1, characterized in that: The step of comparing the training sample after the variable analysis with the test sample and adjusting the similarity between the training sample and the test sample by the preference weight of the training sample comprises: The density ratio of the training samples to the test samples is calculated using transfer learning; the number of the test samples is the same as the number of the training samples, and the density ratio is the preference weight of the training samples during the training process of the patient matching model.

4. The patient record matching method based on the patient primary index according to claim 1, characterized in that: The step of evaluating the effect of the weight and the threshold through the test sample includes: Get user-defined weights and custom thresholds; Generate a custom statistical indicator of the custom weight and the custom threshold on the test sample; Compare the statistical indicators of the patient matching model with the custom statistical indicators; the statistical indicators include: accuracy, percentage of manual participation, recall rate, false positive rate and F1 value.

5. The patient record matching method based on patient primary index according to claim 1, characterized in that: The patient record matching method based on the patient primary index also includes: The results of the effect evaluation are presented visually, and the visual presentation method includes: at least one of text, pictures or tables.

6. The patient record matching method based on the patient primary index according to claim 5, characterized in that: The step of visually presenting the results of the effect evaluation includes: Visual web access is performed for statistical indicator results of the weight and the threshold in the test sample.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the patient record matching method based on the patient master index according to any one of claims 1 to 6 is implemented.

8. An electronic device, characterized in that: include: Processor and memory; The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the electronic device performs the patient record matching method based on a patient master index as described in any one of claims 1 to 6.

Images