Patient nursing risk assessment early warning monitoring method

Abstract

The invention discloses a patient nursing risk assessment early warning monitoring method comprising the following steps: S1, acquiring basic information of a patient object, including diagnosis information and treatment information; s2, performing risk assessment according to the basic information of the patient object to obtain a risk assessment result; and S3, carrying out safety nursing early warning on the patient object according to a risk assessment result. According to the method, the GCN is introduced to extract the semantic similarity and relevance between the diagnosis information and the treatment information, the multi-head self-attention layer and the feedforward network layer are introduced to extract the context between the diagnosis information and the treatment information, and the aggregation operation is utilized to realize the interaction between the two different diagnosis information and treatment information features; the feature representation capability is improved, so that some key aspect information in diagnosis information and treatment information of the patient object is obtained, the nursing risk is evaluated according to a relation prediction result, a corresponding safety nursing plan is formulated, and the safety nursing quality of the patient object is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of risk assessment and early warning, and in particular relates to a patient care risk assessment and early warning monitoring method. Background Art

[0002] Early warning of nursing risks for inpatients is an important part of hospital risk warning. By assessing the health status, disease risk and individual characteristics of patients, a nursing risk early warning mechanism is established to identify potential nursing risks and safety hazards. The purpose of this early warning mechanism is to improve nurses' risk identification and response capabilities and reduce or avoid the occurrence of adverse nursing events. However, the current level of informatization and irregular risk assessment in nursing risk early warning not only affect the effect of safe nursing, but also increase the workload of medical staff. Therefore, how to use information technology and scientific evaluation tools to help nurses identify risks early and improve their ability to control risks is a key topic in nursing risk early warning. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, a patient nursing risk assessment early warning monitoring method provided by the present invention solves the current problems of low information level and non-standard risk assessment in nursing risk early warning.

[0004] In order to achieve the above-mentioned purpose of the invention, the technical solution adopted by the present invention is: a patient care risk assessment early warning monitoring method, comprising the following steps:

[0005] S1. Obtain basic information of the patient, including diagnosis and treatment information;

[0006] S2. Perform risk assessment based on the basic information of the patient and obtain risk assessment results;

[0007] S3. Provide safety care warnings to patients based on risk assessment results.

[0008] Further: S2 comprises the following sub-steps:

[0009] S21, establishing a node feature representation based on the text of the diagnosis information and treatment information of the patient object, and then constructing graph data of the semantic relationship between the diagnosis information and the treatment information based on the node feature representation;

[0010] S22, inputting the graph data of the semantic relationship between the diagnosis information and the treatment information into the GCN, extracting the feature representation of the graph data through the GCN, and obtaining the first text feature representation of the diagnosis information and the treatment information;

[0011] S23, inputting the text of the diagnosis information and the treatment information into the multi-head self-attention layer and the feedforward network layer in sequence to obtain a second text feature representation of the diagnosis information and the treatment information;

[0012] S24, performing feature fusion on the first text feature representation and the second text feature representation of the diagnosis information and the treatment information to obtain a fusion result;

[0013] S25, inputting the fusion result into the Softmax layer to obtain the relationship prediction result between the diagnosis information and the treatment information;

[0014] S26. Evaluate the nursing risk of the patient object based on the relationship prediction results between the diagnosis information and the treatment information to obtain a risk assessment result.

[0015] Further: in said S21, the graph data of the semantic relationship between the diagnosis information and the treatment information includes a node set and an edge set, and the method of constructing the graph data of the semantic relationship between the diagnosis information and the treatment information is specifically:

[0016] The text of diagnosis information and treatment information is split into multiple groups of words, a node set of graph data is constructed based on all the words, and an edge set of graph data is constructed based on the semantic relationship between the words of diagnosis information and treatment information;

[0017] The method for constructing the edge set of graph data is as follows:

[0018] Calculate the point-to-point mutual information between the diagnosis information and treatment information words. If the point-to-point mutual information between the diagnosis information and treatment information words is greater than 0, establish an edge between the diagnosis information and treatment information words.

[0019] Among them, the point-to-point mutual information pmi(x i ;y j ) is expressed as:

[0020]

[0021] Where w(xi,yi) is the word x containing diagnostic information i With treatment information word y j , w(xi) is the number of sliding windows containing diagnostic information words xi, p(yj) is the number of sliding windows containing treatment information words yj, and W is the total number of sliding windows.

[0022] The beneficial effect of the above further scheme is that the point-to-point mutual information can measure the correlation between two words, and the semantic similarity and correlation between the diagnosis information and treatment information words can be described based on the point-to-point mutual information.

[0023] Further: In S22, the method of extracting feature representation of graph data by GCN is specifically as follows:

[0024] The graph data is convolved through the nodes of GCN. The calculation process of a graph convolution operation performed by a node is as follows:

[0025]

[0026]

[0027]

[0028] In the formula, H (l) is the input of the l-th layer GCN, A is the adjacency matrix of the undirected graph, I is the identity matrix, is the intermediate matrix, for The degree matrix of Used to normalize the rows and columns of the matrix, W (l) is the weight matrix of the lth layer, W (l) =R d*h , where h is the output dimension and σ is the nonlinear activation function.

[0029] The beneficial effects of the above further scheme are as follows: GCN uses the structural information of the connection between the edges and nodes of the graph data to propagate and extract features, extract the implicit graph information, and extract the feature representation of the nodes. After the feature extraction of GCN, the semantic information of the diagnosis information and treatment information text can be better captured, which is conducive to the classification of the text and the removal of some redundant information.

[0030] Further: S23 includes the following sub-steps:

[0031] S231, convert the text of diagnosis information and treatment information into a word vector sequence, input it into a multi-head self-attention layer, and obtain a word vector space representation;

[0032]

[0033] Where Q is the query matrix, V is the value matrix, K is the key matrix, and d is the dimension of the key matrix;

[0034] S232. Input the word vector space representation into the feed-forward network layer to obtain a second text feature representation of the diagnosis information and the treatment information.

[0035] The beneficial effects of the above further scheme are: the multi-head self-attention layer is used to extract the meanings of different words while considering the context, and the feedforward network layer is used to further process the output of the multi-head self-attention layer, thereby outputting the contextual semantic features in the text and obtaining a second text feature representation of the diagnostic information and treatment information.

[0036] Further: the S24 is specifically:

[0037] The first text feature representation and the second text feature table of the diagnosis information and the treatment information are sequentially subjected to maximum pooling, concatenation and element-by-element addition operations to obtain a fusion result.

[0038] The beneficial effects of the present invention are as follows: the present invention provides a patient nursing risk assessment, early warning and monitoring method, including, in order to obtain more comprehensive basic information of patient objects and improve the effectiveness of nursing risk assessment and early warning, by introducing GCN to extract the semantic similarity and correlation between diagnostic information and treatment information, introducing a multi-head self-attention layer and a feedforward network layer to extract the context between diagnostic information and treatment information, using aggregation operations to realize the mutual interaction of two different diagnostic information and treatment information features, improving the feature representation capability, thereby obtaining some key aspects of the patient object's diagnostic information and treatment information, evaluating the nursing risk based on the relationship prediction results, and formulating a corresponding safety nursing plan to improve the quality of safe nursing care for the patient object. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flow chart of a patient care risk assessment, early warning and monitoring method of the present invention. DETAILED DESCRIPTION

[0040] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0041] Embodiment 1:

[0042] like Figure 1 As shown, in one embodiment of the present invention, a patient care risk assessment early warning monitoring method includes the following steps:

[0043] S1. Obtain basic information of the patient, including diagnosis and treatment information;

[0044] S2. Perform risk assessment based on the basic information of the patient and obtain risk assessment results;

[0045] S3. Provide safety care warnings to patients based on risk assessment results.

[0046] In this embodiment, the present invention mines diagnostic information and treatment information, and proposes a risk assessment method based on basic information of patient objects, thereby improving the stability and comprehensiveness of nursing risk assessment and early warning.

[0047] The S2 comprises the following sub-steps:

[0048] S21, establishing a node feature representation based on the text of the diagnosis information and treatment information of the patient object, and then constructing graph data of the semantic relationship between the diagnosis information and the treatment information based on the node feature representation;

[0049] S22, inputting the graph data of the semantic relationship between the diagnosis information and the treatment information into the GCN, extracting the feature representation of the graph data through the GCN, and obtaining the first text feature representation of the diagnosis information and the treatment information;

[0050] S23, inputting the text of the diagnosis information and the treatment information into the multi-head self-attention layer and the feedforward network layer in sequence to obtain a second text feature representation of the diagnosis information and the treatment information;

[0051] S24, performing feature fusion on the first text feature representation and the second text feature representation of the diagnosis information and the treatment information to obtain a fusion result;

[0052] S25, inputting the fusion result into the Softmax layer to obtain the relationship prediction result between the diagnosis information and the treatment information;

[0053] S26. Evaluate the nursing risk of the patient object based on the relationship prediction results between the diagnosis information and the treatment information to obtain a risk assessment result.

[0054] In S21, the graph data of the semantic relationship between the diagnosis information and the treatment information includes a node set and an edge set. The method for constructing the graph data of the semantic relationship between the diagnosis information and the treatment information is specifically as follows:

[0055] The text of diagnosis information and treatment information is split into multiple groups of words, a node set of graph data is constructed based on all the words, and an edge set of graph data is constructed based on the semantic relationship between the words of diagnosis information and treatment information;

[0056] In this embodiment, the semantic association information of diagnostic information and treatment information can be described by the edges between nodes in the graph data. For any diagnostic information, its semantic relationship with the corresponding treatment information can be represented by its graph data D = (T, E), where T is a node set, including all words of diagnostic information and treatment information, and E is an edge set, including edges connecting the words of diagnostic information and treatment information, which are used to represent the semantic similarity and association between the two words connected by the edge.

[0057] The method for constructing the edge set of graph data is as follows:

[0058] Calculate the point-to-point mutual information between the diagnosis information and treatment information words. If the point-to-point mutual information between the diagnosis information and treatment information words is greater than 0, establish an edge between the diagnosis information and treatment information words.

[0059] Among them, the point-to-point mutual information pmi(x i ;y j ) is expressed as:

[0060]

[0061] Where w(xi,yi) is the word x containing diagnostic information i With treatment information word y j , w(xi) is the number of sliding windows containing diagnostic information words xi, p(yj) is the number of sliding windows containing treatment information words yj, and W is the total number of sliding windows.

[0062] In this embodiment, the point-to-point mutual information can measure the correlation between two words, and the semantic similarity and correlation between the diagnosis information and treatment information words can be described based on the point-to-point mutual information.

[0063] In S22, the method for GCN to extract feature representation of graph data is specifically as follows:

[0064] The graph data is convolved through the nodes of GCN. The calculation process of a graph convolution operation performed by a node is as follows:

[0065]

[0066]

[0067]

[0068] In the formula, H (l) is the input of the l-th layer GCN, A is the adjacency matrix of the undirected graph, I is the identity matrix, is the intermediate matrix, for The degree matrix of Used to normalize the rows and columns of the matrix, W (l) is the weight matrix of the lth layer, W (l) =R d*h , where h is the output dimension and σ is the nonlinear activation function.

[0069] In this embodiment, GCN uses the structural information of the connection between graph data edges and nodes to propagate and extract features, extract implicit graph information, and extract the feature representation of the nodes. After feature extraction by GCN, the semantic information of the diagnosis information and treatment information text can be better captured, which is conducive to text classification and removal of some redundant information.

[0070] The S23 comprises the following sub-steps:

[0071] S231, convert the text of diagnosis information and treatment information into a word vector sequence, input it into a multi-head self-attention layer, and obtain a word vector space representation;

[0072]

[0073] Where Q is the query matrix, V is the value matrix, K is the key matrix, and d is the dimension of the key matrix;

[0074] S232. Input the word vector space representation into the feed-forward network layer to obtain a second text feature representation of the diagnosis information and the treatment information.

[0075] In this embodiment, the multi-head self-attention layer is used to extract the meaning of different words while considering the context, and the feedforward network layer is used to further process the output of the multi-head self-attention layer, thereby outputting the contextual semantic features in the text and obtaining a second text feature representation of the diagnostic information and treatment information.

[0076] The S24 is specifically:

[0077] The first text feature representation and the second text feature table of the diagnosis information and the treatment information are sequentially subjected to maximum pooling, concatenation and element-by-element addition operations to obtain a fusion result.

[0078] In this embodiment, by performing an aggregation operation on the first text feature representation and the second text feature representation of the diagnosis information and the treatment information, the accuracy and performance of classification can be improved, and the robustness and stability of the relationship prediction result in the next step can be improved.

[0079] In S25, based on the prediction results of the relationship between the diagnostic information and the treatment information, the physical condition, potential risk factors and nursing needs of the patient can be obtained to help analyze whether the patient has nursing risks, provide more comprehensive text semantic features for subsequent nursing risk assessment, and improve the effectiveness of the risk assessment results.

[0080] In S26, key information between the diagnosis information and treatment information of the patient object can be obtained according to the relationship prediction result, which can help the nurse to perform accurate risk assessment on the patient object.

[0081] In S3, the method for providing a safety nursing warning to the patient object is specifically as follows:

[0082] Based on the risk assessment of the patient's diagnostic information and treatment information, combined with the patient's medical records, examination and test results, doctor's orders, surgical records, nursing records, etc., a safe care plan is formulated to improve the quality of care and reduce medical errors and hidden dangers.

[0083] The beneficial effects of the present invention are as follows: the present invention provides a patient nursing risk assessment, early warning and monitoring method, including, in order to obtain more comprehensive basic information of patient objects and improve the effectiveness of nursing risk assessment and early warning, by introducing GCN to extract the semantic similarity and correlation between diagnostic information and treatment information, introducing a multi-head self-attention layer and a feedforward network layer to extract the context between diagnostic information and treatment information, using aggregation operations to realize the mutual interaction of two different diagnostic information and treatment information features, improving the feature representation capability, thereby obtaining some key aspects of the patient object's diagnostic information and treatment information, evaluating the nursing risk based on the relationship prediction results, and formulating a corresponding safety nursing plan to improve the quality of safe nursing care for the patient object.

[0084] In the description of the present invention, it is necessary to understand that the orientation or positional relationship indicated by the terms "center", "thickness", "upper", "lower", "horizontal", "top", "bottom", "inner", "outer", "radial", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", and "third" are used only for descriptive purposes, and cannot be understood as indicating or implying the relative importance or the number of implicitly specified technical features. Therefore, the features defined by "first", "second", and "third" may explicitly or implicitly include one or more of the features.

Claims

1. A patient care risk assessment early warning monitoring method, characterized in that: The following steps are involved: S1. Obtain basic information of the patient, including diagnosis and treatment information; S2. Perform risk assessment based on the basic information of the patient and obtain risk assessment results; S3. Provide safety care warnings to patients based on risk assessment results.

2. The patient care risk assessment, early warning and monitoring method according to claim 1, characterized in that: The S2 comprises the following sub-steps: S21, establishing a node feature representation based on the text of the diagnosis information and treatment information of the patient object, and then constructing graph data of the semantic relationship between the diagnosis information and the treatment information based on the node feature representation; S22, inputting the graph data of the semantic relationship between the diagnosis information and the treatment information into the GCN, extracting the feature representation of the graph data through the GCN, and obtaining the first text feature representation of the diagnosis information and the treatment information; S23, inputting the text of the diagnosis information and the treatment information into the multi-head self-attention layer and the feedforward network layer in sequence to obtain a second text feature representation of the diagnosis information and the treatment information; S24, performing feature fusion on the first text feature representation and the second text feature representation of the diagnosis information and the treatment information to obtain a fusion result; S25, inputting the fusion result into the Softmax layer to obtain the relationship prediction result between the diagnosis information and the treatment information; S26. Evaluate the nursing risk of the patient object based on the relationship prediction results between the diagnosis information and the treatment information to obtain a risk assessment result.

3. The patient care risk assessment, early warning and monitoring method according to claim 2, characterized in that: In S21, the graph data of the semantic relationship between the diagnosis information and the treatment information includes a node set and an edge set. The method for constructing the graph data of the semantic relationship between the diagnosis information and the treatment information is specifically as follows: The text of diagnosis information and treatment information is split into multiple groups of words, a node set of graph data is constructed based on all the words, and an edge set of graph data is constructed based on the semantic relationship between the words of diagnosis information and treatment information; The method for constructing the edge set of graph data is as follows: Calculate the point-to-point mutual information between the diagnosis information and treatment information words. If the point-to-point mutual information between the diagnosis information and treatment information words is greater than 0, establish an edge between the diagnosis information and treatment information words. Among them, the point-to-point mutual information pmi(x i ;y j ) is expressed as: Where w(xi,yi) is the word x containing diagnostic information i With treatment information word y j , w(xi) is the number of sliding windows containing diagnostic information words xi, p(yj) is the number of sliding windows containing treatment information words yj, and W is the total number of sliding windows.

4. The patient care risk assessment and early warning monitoring method according to claim 2 is characterized in that: In S22, the method for GCN to extract feature representation of graph data is specifically as follows: The graph data is convolved through the nodes of GCN. The calculation process of a graph convolution operation performed by a node is as follows: In the formula, H (l) is the input of the l-th layer GCN, A is the adjacency matrix of the undirected graph, I is the identity matrix, is the intermediate matrix, for The degree matrix of Used to normalize the rows and columns of the matrix, W (l) is the weight matrix of the lth layer, W (l) =R d*h , where h is the output dimension and σ is the nonlinear activation function.

5. The patient care risk assessment, early warning and monitoring method according to claim 2, characterized in that: The S23 comprises the following sub-steps: S231, convert the text of diagnosis information and treatment information into a word vector sequence, input it into a multi-head self-attention layer, and obtain a word vector space representation; Where Q is the query matrix, V is the value matrix, K is the key matrix, and d is the dimension of the key matrix; S232. Input the word vector space representation into the feed-forward network layer to obtain a second text feature representation of the diagnosis information and the treatment information.

6. The patient care risk assessment, early warning and monitoring method according to claim 2, characterized in that: The S24 is specifically: The first text feature representation and the second text feature table of the diagnosis information and the treatment information are sequentially subjected to maximum pooling, concatenation and element-by-element addition operations to obtain a fusion result.

Images