A hospital examination test name alignment method based on community discovery

Abstract

A hospital examination test name alignment method based on community discovery, a patient's disease diagnosis, operation and examination test in a clinical record are constituted into a graph structure, and then a community discovery algorithm is used to cluster the graph structure into different communities, the edges in the community are closely connected and the edges between different communities are sparse, the internal of a community all belong to disease diagnosis, operation or examination test which are closely related to each other, so when the examination test name alignment is performed, the alignment object is only searched from the community where the examination test is located, thereby reducing the deviation of the examination test name alignment.

Description

Technical Field

[0001] The present invention relates to the technical field of medical big data, and in particular to a method for aligning hospital examination and test names based on community discovery. Background Art

[0002] Examinations and tests are fundamental to disease diagnosis, but different hospitals may use different names for the same test. For example, at Hospital A, a test is called "Vertebral Artery Color Doppler Ultrasound," while at Hospital B, it's called "Color Doppler Ultrasound (Conical Artery)." While these two tests have different names, the underlying content, method, location, and purpose of the tests are the same. This diversity in test names poses challenges for statistical analysis of medical data and the development of medical big data platforms. Traditional string alignment methods based on edit distance have poor generalization for rigorous medical and health data. For example, when attempting to align "Vertebral Artery Color Doppler Ultrasound" among a large number of different test names using edit distance, the alignment target found is "Extremity Artery Color Doppler Ultrasound" rather than "Extremity Artery Color Doppler Ultrasound." This alignment error occurs because the edit distance between "Vertebral Artery Color Doppler Ultrasound" and "Extremity Artery Color Doppler Ultrasound" is smaller. Although the two names differ by only two characters, the locations examined are completely different. Summary of the Invention

[0003] In order to overcome the deficiencies of the above technologies, the present invention provides a method for reducing the deviation of checking and verifying name alignment.

[0004] The technical solution adopted by the present invention to overcome the technical problems is:

[0005] A method for aligning hospital examination and test names based on community discovery includes the following steps:

[0006] a) Obtain an electronic medical record dataset from N hospitals. The dataset contains n electronic medical records. Each electronic medical record includes the disease code and surgery code on the front page of the medical record, as well as the detailed names of the examination and test items for the patient's admission. The number of disease code categories in the electronic medical record dataset is d, the number of surgery code categories is p, and the number of examination and test item categories is τ;

[0007] b) Create index IDs for the categories of disease codes, surgical codes, and examination and test item names in the electronic medical record dataset;

[0008] c) Initialize the adjacency matrix A, is a real number space, m is the dimension, and the value of the i-th row and j-th column in the adjacency matrix A is A ij, i ∈ {1, 2,..., m}, j ∈ {1, 2,..., m};

[0009] d) The subset corresponding to the top P% largest values in the adjacency matrix A is Using the subset An adjacency matrix A' is constructed,

[0010] e) The adjacency matrix A' is decomposed to obtain m eigenvalues, and the absolute values of each eigenvalue are arranged in descending order to obtain the eigenvectors Λ of the adjacency matrix A', λ i is the ith eigenvalue, i ∈ {1,..., m}, and the eigenvector corresponding to the ith eigenvalue λ i is v(λ i ), and the eigenvector corresponding to Λ is The second derivative Δ 2 of the eigenvalues in the eigenvector Λ is calculated Where The maximum value of the second derivative Δ 2 is The subscript index corresponding to the maximum value is k, k ∈ {1,..., m-2};

[0011] f) An m × (m-1) matrix G is calculated, and the K-means algorithm is used to cluster the m dimension of the matrix G, and the matrix G is clustered into k+2 communities to obtain the clustering result C, C = (c1, c2,..., c r ,..., c k+2 ), where c r is the rth community, r ∈ {1,..., k+2}, and the rth community c r contains the encoded category, the surgical code category, and the category of the examination item name Where is the encoded category, the surgical code category, and the category of the examination item name

[0012] g) An examination name that needs to be aligned is obtained, and the community in which the examination name is located is found The category of the examination name contained in the community is found to obtain the set Z of the examination name Where is the examination name |z| is the number of all examination names in the community , and the examination name that needs to be aligned is calculated the levenshtein distance between the i-th element of the index ids and the j-th element of the index ids in the set Z, and the levenshtein distance between the i-th element of the index ids and the j-th element of the index ids in the set Z is the minimum the levenshtein distance between the i-th element of the index ids and the j-th element of the index ids in the set Z j aligning z j with .

[0013] Preferably, the disease codes and the operation codes in step a) are the standard categories of diseases and the standard categories of operations after the ICD codes of the international disease classification are standardized.

[0014] Further, in step b), the index ids = (1, 2, …, m) is used, where m = d + p + τ, ids[1:d] is the first element to the d-th element of the index ids, which represents the index of the category of the disease codes, ids[d+1:d+p] is the (d+1)-th element to the (d+p)-th element of the index ids, which represents the index of the category of the operation codes, and ids[d+p+1:d+p+τ] is the (d+p+1)-th element to the (d+p+τ)-th element of the index ids, which represents the index of the category of the examination test item name.

[0015] Further, in step c), the value of the i-th row and the j-th column of the adjacency matrix A is calculated by the formula , ij , #ij is the number of times that the disease code category, the operation code category, and the examination test item name category corresponding to the i-th index of the index ids and the disease code category, the operation code category, and the examination test item name category corresponding to the j-th index of the index ids appear together in the same electronic medical record, #i is the number of times that the disease code category, the operation code category, and the examination test item name category corresponding to the i-th index of the index ids appear in the electronic medical record dataset, #j is the number of times that the disease code category, the operation code category, and the examination test item name category corresponding to the j-th index of the index ids appear in the electronic medical record dataset.

[0016] Preferably, in step d), P is 3.

[0017] Further, in step d), the adjacency matrix A′ is calculated by the formula i ′ j , , where A i ′ jis the undirected edge formed by connecting the disease code category, surgical code category, and inspection and test item name category corresponding to the i-th node in the adjacency matrix A′ with the disease code category, surgical code category, and inspection and test item name category corresponding to the j-th node. j ' i It is an undirected edge formed by connecting the disease code category, surgery code category, and examination and inspection item name category corresponding to the j-th node in the adjacency matrix A′ with the disease code category, surgery code category, and examination and inspection item name category corresponding to the i-th node.

[0018] Furthermore, in step f), the formula The matrix G is calculated. The beneficial effects of the present invention are as follows: a patient's disease diagnosis, surgical operation, and examination and test in the clinical record are formed into a graph structure, and then the graph structure is clustered into different communities using a community discovery algorithm, where the edges within the community are closely connected and the edges between different communities are sparsely connected. The interior of a community all belongs to disease diagnoses, surgical operations, or examinations that are closely related to each other, so that when aligning the names of examinations and tests, alignment objects are only found from the community where the examination and test is located, thereby reducing the deviation of the alignment of the examination and test names. DETAILED DESCRIPTION

[0019] The present invention will be further described below.

[0020] A method for aligning hospital examination and test names based on community discovery includes the following steps:

[0021] a) Obtain an electronic medical record dataset from N hospitals. This dataset contains n electronic medical records. Each electronic medical record includes the disease code and surgery code on the front page of the medical record, as well as a detailed list of the names of the examinations and tests performed on the patient during that admission. The disease code indicates the disease treated during the patient's admission, and the surgery code indicates the surgery performed on the patient during that admission. The number of disease code categories in the electronic medical record dataset is d, the number of surgery code categories is p, and the number of examination and test item categories is τ. The purpose of this patent is to align the names of examination and test items across different hospitals.

[0022] b) Create index IDs for the categories of disease codes, surgical codes, and examination and inspection item names in the electronic medical record dataset.

[0023] c) Initialize the adjacency matrix A, is a real number space, m is the dimension, and the value of the i-th row and j-th column in the adjacency matrix A is A ij , i∈{1,2,...,m}, j∈{1,2,...,m}.

[0024] d) The subset corresponding to the largest values ​​in the first P% of the adjacency matrix A is Utilize subsets Construct the adjacency matrix A′, The adjacency matrix A′ represents an undirected graph structure (or network structure) with m nodes.

[0025] e) Perform eigendecomposition on the adjacency matrix A′ to obtain m eigenvalues. Arrange the absolute values ​​of each eigenvalue in descending order to obtain the eigenvector Λ of the adjacency matrix A′. λ i is the i-th eigenvalue, i∈{1,...,m}, the i-th eigenvalue λ i The eigenvector of is v(λ i ), the eigenvector corresponding to Λ is Calculate the second-order derivative Δ of the eigenvalue in the eigenvector Λ 2 , in Second-order derivative Δ 2 The largest value in The corresponding subscript index is k, k∈{1,...,m-2}.

[0026] f) Calculate an m×(m-1)-dimensional matrix G, use the K-means algorithm to cluster the matrix G in the m dimension of the matrix G, cluster the matrix G into k+2 communities, and obtain the clustering result C, C = (c1, c2, ..., c r ,...,c k+2 ), where c r is the rth community, r∈{1,...,k+2}, the rth community c r The categories of codes, surgical codes, and inspection item names contained in in For the The categories of individual codes, surgical codes, and examination and testing item names,

[0027] g) Get the name of a check that needs to be aligned and find the community where the check name is located From the community Find the categories of inspection and test names contained in the community and get the set Z of inspection and test names. in For the Name of inspection test, |z| is a community The number of all inspection and test names in the , calculate the inspection and test names that need to be aligned The levenshtein distance between the other examination test names in the set Z and the levenshtein distance between the first examination test name in the set Z is the smallest element z The levenshtein distance between the other examination test names in the set Z and the levenshtein distance between the first examination test name in the set Z is the smallest element z The levenshtein distance between the other examination test names in the set Z and the levenshtein distance between the first examination test name in the set Z is the smallest element z j The levenshtein distance between the other examination test names in the set Z and the levenshtein distance between the first examination test name in the set Z is the smallest element z j The levenshtein distance between the other examination test names in the set Z and the levenshtein distance between the first examination test name in the set Z is the smallest element z The levenshtein distance between the other examination test names in the set Z and the levenshtein distance between the first examination test name in the set Z is the smallest element z j The levenshtein distance between the other examination test names in the set Z and the levenshtein distance between the first examination test name in the set Z is the smallest element z The levenshtein distance between the other examination test names in the set Z and the levenshtein distance between the first examination test name in the set Z is the smallest element z

[0028] The levenshtein distance between the other examination test names in the set Z and the levenshtein distance between the first examination test name in the set Z is the smallest element z

[0029] The levenshtein distance between the other examination test names in the set Z and the levenshtein distance between the first examination test name in the set Z is the smallest element z

[0030]

[0031] The levenshtein distance between the other examination test names in the set Z and the levenshtein distance between the first examination test name in the set Z is the smallest element z

[0032] Embodiment 1:

[0033] The disease code and the operation code in step a) are the standard categories of diseases and the standard categories of operations after the ICD code of the international disease classification is standardized.

[0034] Embodiment 2:

[0035] In step b), the index ids=(1, 2,..., m), where m=d+p+τ, ids[1:d] is the first element to the dth element in the index ids, which represents the index of the category of the disease code, ids[d+1:d+p] is the (d+1)th element to the (d+p)th element in the index ids, which represents the index of the category of the operation code, and ids[d+p+1:d+p+τ] is the (d+p+1)th element to the (d+p+τ)th element in the index ids, which represents the index of the category of the examination test item name.

[0036] Embodiment 3:

[0037] The value of the i-th row and j-th column in the adjacency matrix A is calculated by the formula Aij= #ij / (#i * #j) ij , #ij is the number of times that the disease code category, surgery code category, and examination item name category corresponding to the i-th index in the index ids and the disease code category, surgery code category, and examination item name category corresponding to the j-th index co-occur in the same electronic medical record, #i is the number of times that the disease code category, surgery code category, and examination item name category corresponding to the i-th index in the index ids appears in the electronic medical record dataset, #j is the number of times that the disease code category, surgery code category, and examination item name category corresponding to the j-th index in the index ids appears in the electronic medical record dataset.

[0038] Example 4:

[0039] In step d), P takes the value of 3.

[0040] Example 5:

[0041] In step d), the adjacency matrix A' is calculated by the formula A i ′ j , The value of the i-th node and j-th node in the adjacency matrix A' is calculated by the formula i ′ j A'ij is the undirected edge formed by connecting the disease code category, surgery code category, and examination item name category corresponding to the i-th node in the adjacency matrix A' and the disease code category, surgery code category, and examination item name category corresponding to the j-th node, A j ′ i A'ji is the undirected edge formed by connecting the disease code category, surgery code category, and examination item name category corresponding to the j-th node in the adjacency matrix A' and the disease code category, surgery code category, and examination item name category corresponding to the i-th node. If A i ′ j = 1, then the disease code category, surgery code category, and examination item name category corresponding to the i-th node and the disease code category, surgery code category, and examination item name category corresponding to the j-th node are connected by an undirected edge, and in the adjacency matrix A' representing the graph structure, if A i ′ j and A j ′ i are both 1, then it indicates that the disease code category, surgery code category, and examination item name category corresponding to the i-th node and the disease code category, surgery code category, and examination item name category corresponding to the j-th node are connected by an undirected edge, and in the adjacency matrix A' representing the graph structure,i ' j ,A j ' i =0, the disease code category, surgery code category, and inspection and test item name category corresponding to the i-th node have no edge connection with the disease code category, surgery code category, and inspection and test item name category corresponding to the j-th node.

[0042] Example 6:

[0043] In step f), the formula The matrix G is calculated.

[0044] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A hospital examination test name alignment method based on community discovery, characterized in that, comprising the steps of: a) obtaining an electronic medical record dataset of N hospitals, the electronic medical record dataset containing a plurality of electronic medical records, each electronic medical record including disease codes and surgery codes in a medical record home page and details of examination and test item names for a patient's admission, the number of categories of disease codes in the electronic medical record dataset being , the number of categories of surgery codes being , and the number of categories of examination and test item names being ; b) making an index of the categories of disease codes, the categories of procedure codes, and the categories of examination test item names in the electronic medical record data set ; c) initializing an adjacency matrix , , is a real space, is a dimension, an adjacency matrix is the value in the i-th row and j-th column of the adjacency matrix , , ;​​ d) an adjacency matrix The subset corresponding to the largest value of P is , the subset is used to construct the adjacency matrix , ; e) the adjacency matrix is decomposed to obtain characteristic values, and the absolute values of each characteristic value are arranged in descending order to obtain an adjacency matrix characteristic vector , , is the first characteristic value, is the first characteristic value is the characteristic vector of the first characteristic value, is the characteristic vector corresponding to the first characteristic value, is the second derivative of the characteristic value in the characteristic vector , is the second derivative of the characteristic value in the characteristic vector , is the maximum value of the second derivative , is the index corresponding to the maximum value of the second derivative , ; f) calculate a dimensional matrix , use the K-means algorithm to cluster the matrix In the matrix The dimension of , the matrix is clustered into communities, and the clustering result , , where is the th community, , the th community The category of disease code, the category of surgery code or the category of examination item name contained in the community is represented as , where is the th category of disease code, the category of surgery code or the category of examination item name, ; g) Get the name of an inspection project that needs to be aligned and find the community where the inspection project name is located , from the community Find the categories of inspection and testing item names contained in the community and get the set of inspection and testing item names , ,in For the Name of inspection item, , For the community The number of all inspection and test item names in the , calculate the inspection and test item names that need to be aligned With collection Levenshtein distance of other inspection items in the set Middle and First Name of inspection item The element with the smallest Levenshtein distance is ,Will and Perform alignment; The value of the (i, j)th element of the adjacency matrix A is calculated by the formula The value of the (i, j)th element of the adjacency matrix A is calculated by the formula The value of the (i, j)th element of the adjacency matrix A is calculated by the formula The value of the (i, j)th element of the adjacency matrix A is calculated by the formula The value of the (i, j)th element of the adjacency matrix A is calculated by the formula , , The value of the (i, j)th element of the adjacency matrix A is calculated by the formula The value of the (i, j)th element of the adjacency matrix A is calculated by the formula The value of the (i, j)th element of the adjacency matrix A is calculated by the formula The value of the (i, j)th element of the adjacency matrix A is calculated by the formula , The value of the (i, j)th element of the adjacency matrix A is calculated by the formula The value of the (i, j)th element of the adjacency matrix A is calculated by the formula The value of the (i, j)th element of the adjacency matrix A is calculated by the formula , The value of the (i, j)th element of the adjacency matrix A is calculated by the formula The value of the (i, j)th element of the adjacency matrix A is calculated by the formula The value of the (i, j)th element of the adjacency matrix A is calculated by the formula 2. The community discovery based hospital examination test name alignment method of claim 1, wherein: The disease code and the surgery code in step a) are the standard categories of the disease and the standard categories of the surgery after the ICD code using the International Classification of Diseases is standardized.

3. The community discovery based hospital examination test name alignment method of claim 1, wherein: Index in step b) ,in , For index The first element to the elements, which represent the index of the disease code category, For index Middle Elements to elements, which represent the index of the category of surgical coding, For index Middle Elements to An element that represents the index of the category of the inspection item name.

4. The community discovery based hospital examination test name alignment method of claim 1, wherein: The P value in step d) is 3.

5. The community discovery based hospital examination test name alignment method of claim 1, wherein: The adjacency matrix A is calculated by the formula The adjacency matrix A is calculated by the formula wherein is the adjacency matrix A in step d) of the formula is the disease code category, the surgery code category or the examination item name category corresponding to the i-th node in the adjacency matrix A is the disease code category, the surgery code category or the examination item name category corresponding to the i-th node in the adjacency matrix A is the disease code category, the surgery code category or the examination item name category corresponding to the i-th node in the adjacency matrix A is the disease code category, the surgery code category or the examination item name category corresponding to the i-th node in the adjacency matrix A is the disease code category, the surgery code category or the examination item name category corresponding to the i-th node in the adjacency matrix A is the disease code category, the surgery code category or the examination item name category corresponding to the i-th node in the adjacency matrix A is the disease code category, the surgery code category or the examination item name category corresponding to the i-th node in the adjacency matrix A 6. The community discovery based hospital examination test name alignment method of claim 1, wherein: The matrix is calculated in step f) by the formula .​

Images