A system to reduce nurses' medication errors

Abstract

This invention proposes a system to reduce medication errors by nurses, comprising: a smartphone with an installed application, a pharmacy management system, a web server, an NFC tag affixed to the medication packaging, and an NFC reader for the pharmacy management system. The pharmacy management system receives patient information sent by the patient, verifies the patient's identity with the web server, and obtains the necessary information for medication dispensing. The web server verifies the patient's identity; if successful, it sends back information, and the nurse scans the medication. The NFC tag affixed to the medication packaging contains the medication name and ID. The pharmacy management system, after scanning the NFC tag, verifies the medication ID by matching it with the ID in the prescription information. This system makes prescription information readily available to patients, provides high-quality service by minimizing medication errors, ensures the security and privacy of patients' sensitive information, and reduces the time required to prepare prescriptions in advance.

Description

Technical Field

[0001] This invention relates to the field of nursing, and more specifically, to a system for reducing medication errors by nurses. Background Technology

[0002] Medication errors are the most common medical errors in healthcare settings. Medication delivery is a complex, multi-stage process involving multiple healthcare professionals. Medication errors can occur at any step of the process, with 38% of errors occurring during the administration phase. Nurses spend approximately 40% of the time administering medications, acting as the last line of defense to prevent errors before they reach the patient. In some cases, nurses may be witnesses or submitters of medication errors, and most hospitals rely on nurses to report them. However, due to factors such as heavy workloads, misunderstandings of physician handwriting, or administering the wrong medication to the wrong patient, medication errors remain difficult to prevent. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a system for reducing medication errors by nurses. This system reduces the time required to prepare prescriptions in advance and can quickly identify abnormal values ​​in prescription drugs, thereby preventing medication errors.

[0004] The present invention proposes a system for reducing nurses' medication errors, comprising: a smartphone with an application installed, a pharmacy management system, a web server, an NFC tag affixed to the medicine packaging, and an NFC reader for the pharmacy management system;

[0005] Among them, a fingerprint scanner is embedded in the smartphone, and the patient uses the fingerprint scanner embedded in the smartphone to verify his identity;

[0006] The smartphone application generates a timestamp, and the timestamp value will be sent to the web server.

[0007] The application will request the active prescription ID from the web server. The application will then use the selected prescription ID, patient ID, and timestamp to generate patient information and send it to the web server.

[0008] The pharmacy management system receives patient information sent by patients, verifies their identity with the web server, and obtains the information required for dispensing medication.

[0009] The web server verifies the identity. If the verification is successful, the web server will send back information. The nurse scans the medicine. The medicine packaging has an NFC tag, which contains the medicine name and ID.

[0010] The pharmacy management system, after acquiring the scanned NFC tag, verifies the drug ID by matching it with the ID in the prescription information. If the match is successful, the nurse will dispense the medication.

[0011] The information required for dispensing medication includes the patient's name, medication ID, and prescription ID.

[0012] Upon first use of the application, patients are required to set up their fingerprints. Without a fingerprint, patients will not be able to access the application or transmit any information. This verification step helps verify the patient's medication dispensing process and helps protect the privacy of sensitive patient information. The active prescription ID is always stored on the web server and is not stored on the patient's mobile phone.

[0013] Thoroughly reviewing prescriptions is the final step in the medication dispensing process, extracting relevant data to prevent medication errors. The processed data contains nearly a hundred variables. This type of multivariate data is often referred to as high-dimensional data, which presents challenges for interpretation and analysis due to the curse of dimensionality, data redundancy, and noise. Any model directly built on high-dimensional data typically exhibits poor generalization performance, limiting analytical models and making it difficult to create visualizations of multidimensional data. Visualization plays a crucial role in guiding hospital management and improving nurses' medication error reporting. To improve data interpretability and achieve comprehensive visualization, appropriate variable selection, reducing data dimensionality, and identifying the optimal predictor variables are essential.

[0014] A fast parallel variable selection model is proposed to identify the best predictor of nurse-reported medication errors by using Extreme Learning Machines (ELMs) incorporated into the packaged variable selection mechanism. ELMs are accurate, extremely fast, and non-linear predictive models. During variable selection, variables are selected from the R-values ​​of each ELM model. 2 Value evaluation. The hidden layer weights of the parallelized extreme learning machine are randomly initialized and can be fixed without iterative adjustment. Common activation functions are applied to the hidden neurons. The only parameter learned in the parallelized extreme learning machine is the weights between the hidden and output layers. The parameters of the hidden neurons do not depend on the training data, allowing the parallelized extreme learning machine to reach an approximate optimal generalization bound.

[0015] Parallelized Extreme Learning Machines (PEMs) are used for the rapid training of single-hidden-layer feedforward networks. A single-hidden-layer feedforward neural network has three layers of neurons: the input layer provides data features and does not perform computation; the hidden layer is the only layer that undergoes a non-linear transformation; and the output layer is linear, without any transformation function or bias. The training of a parallelized extreme learning machine consists of two stages: First, random feature mapping. In this stage, the input data is randomly projected into a new space. The parallelized extreme learning machine randomly initializes the input layer weights W∈R. d*L And bias b∈R L , where L is the number of hidden neurons. Input data x∈R d The activation function is nonlinearly mapped to the hidden neuron i, as shown in Equation (1):

[0016] h i (x)=φ(w i x, b i (1)

[0017] Among them, w i ∈R d b∈R It is a non-linear piecewise continuous function. The hidden layer is not limited to having only one type of transformation function in the neurons. The number of linear neurons is equal to the number of data features, and each of these neurons replicates the corresponding feature.

[0018] The second stage of parallelized extreme learning machine is to solve the linear system and find the optimal β that minimizes the following cost function, as shown in Equation (2):

[0019]

[0020] Where H is the output of the hidden layer, as shown in formula (3):

[0021]

[0022] Where Y represents all outcome variables, Y i Let X be the outcome variable i, and X be the predictor variable for all samples. i Let S be the i-th sample of X. k Let k variables be randomly selected from X, W be the input layer weights of the parallelized extreme learning machine, b be the input layer bias of the parallelized extreme learning machine, L be the number of hidden neurons in the parallelized extreme learning machine, h be the output of the hidden neurons in the parallelized extreme learning machine, φ be the nonlinear transformation function in the parallelized extreme learning machine, β be the output layer weights of the parallelized extreme learning machine, and c be the weights of the output layer of the parallelized extreme learning machine. s Let α be the best matching unit in the self-organizing map, and σ be the learning rate in the self-organizing map. λ Let be the neighborhood function in the self-organizing map, and d be the distance function in the self-organizing map.

[0023] The cost function itself is the mean squared error between the approximation of the parallelized extreme learning machine and the true target value Y. The optimal solution of the cost function β* is shown in equation (4):

[0024] β*=H + Y (4)

[0025] Among them, H + It is a pseudo-inverse, and the calculation formula is:

[0026] H + =(H T H) -1 H T(5)

[0027] The implementation of the pseudoinverse includes a small regularization term H. + =(H T H+αI) -1 H T .

[0028] R 2 The value serves as an accurate measure of the performance of variable selection. Due to the different scenarios of outcomes, the relationships between the predictor variables and each outcome variable are evaluated separately, and R0 is calculated. 2 The values ​​are divided into three prediction results for training the parallelized extreme learning machine: (X,Y1), (X,Y2), and (X,Y3).

[0029] R is calculated using the mean squared error of the parallelized extreme learning machine model. 2 The value is shown in formula (6):

[0030]

[0031] The mean squared error value is rewritten as shown in formula (7):

[0032]

[0033] In the formula, Y is the output of the parallelized extreme learning machine model. i The estimate of R. 2 The higher the value, the better the prediction. The more accurate, the better. 2 The value is used to evaluate the performance of the selected variable.

[0034] In the selection of packaging variables, multiple parallelized extreme learning machines were constructed in parallel, each evaluating k variables randomly selected from X. R0 was computed. 2 The value is used to measure the predictive performance of the selected k variables. R 2 The larger the value, the better the prediction from these selected variables, and the larger the R-value. 2 A variable whose value is [value].

[0035] The variable selection process includes the following steps:

[0036] S1: Random Variable Selection: From the original dataset X, randomly select k variables to create a subset of the original data: S k Initially, k = 1, and the current optimal variable set S k * =S k The final optimal set of variables S k ** =S k ;

[0037] S2: Parallelized Extreme Learning Machine Training. The parallelized extreme learning machine is built on the selected data S. k and outcome variable Y i Based on this, calculate R 2 value;

[0038] S3: Update the current best set of predictors. If the current R... 2 If the value is higher than the previous calculation, then update S. k * =S k Otherwise, S k Keep it the same;

[0039] S4: Iterative Update: Repeat iterative steps S61 to S63 until the stopping criterion is met. The stopping criterion is defined by the maximum number of iterations. If the maximum number of iterations is reached, the final optimal variable set S is obtained. k ** By S k * renew;

[0040] S5: Increase the value of k, increase the number of variables selected by one, and start again from step S61.

[0041] Self-organizing maps (SOMs) are used to perform visualization after determining the highest predicted value. SOMs reveal the non-linear relationship between predictors and outcomes, using a predefined two-dimensional grid to capture the topological structure of high-dimensional data. Each point on the grid is assigned a weight, with each point having randomly initialized weights. After numerous iterations, these weights are updated to reflect the position of the point in the original data space. In the iterative algorithm, unit c... s For s = [1, ..., N], where N is the number of points on the two-dimensional grid, it is updated using the rule shown in Equation (8):

[0042] c s ←c s +ασ λ (r, s)(x) i -c s (8)

[0043] Where, x i Let α be the i-th data point, α be the learning rate between 0 and 1, and σ be the learning rate between 0 and 1. λ This is a neighborhood function; it returns 0 for non-neighborhoods and other non-zero values ​​for valid neighborhoods. Furthermore, d is a distance function, r = argmin{d(x...} i ,c s )}.

[0044] According to the self-organizing map algorithm, each point c on the two-dimensional grid... ss = [1,…,N] represents a set of points in a multidimensional data space. s A self-organizing map (SOM) is the optimal matching unit of a set of points in the original data space, performing discrete nonlinear dimensionality reduction. To understand the visualization, the SOM is converted into a heatmap using color, showing the importance of a given variable.

[0045] After extracting given variables, the system automatically learns thresholds between normal and abnormal dosages for each variable in e-prescriptions using previous prescriptions to detect potential outliers. Machine learning eliminates the need for extensive preprocessing, thus reducing computational costs. A density-distance centrality screening method is used; outlier detection involves identifying low-density, low-similarity prescription groups and ranking prescriptions based on their centrality indices. Overdosage or underdosage prescriptions are those with centrality scores lower than the average centrality index for each drug.

[0046] Prescription outliers are calculated by representing each prescription in a vector space, fitting each prescription drug into a two-dimensional vector using dosage and daily frequency. The relationships between prescriptions are represented as a graph, where vertices are prescriptions and edges are based on the similarity between a pair of prescriptions. A similarity function is defined as the pairwise similarity between two-dimensional vectors composed of dosage and frequency. Pairwise similarity accepts any pairwise measure; a normal prescription has a high centrality index, similar to many other prescriptions. To calculate the centrality density of each prescription, the weight of each edge is calculated using extension, weighted links, and weighted nodes to calculate the centrality score of the prescription. The weighting function is given by Equation (9):

[0047]

[0048] Among them, B u It is the set of all neighborhoods u contained in N. u Let W(v, u) represent the number of neighbors of u, and W(v, u) be the weight of the external links from v to u. The more links a prescription drug has with other prescription drugs that are highly similar to it, the more representative it is in prescription drug distribution.

[0049] This invention provides a system for reducing medication errors by nurses, making prescription information readily available to patients, providing high-quality service by minimizing medication errors, ensuring the security and privacy of patients' sensitive information, and reducing the time required to prepare prescriptions in advance. Furthermore, by combining parallelized extreme learning machines and self-organizing maps to extract relevant variables from prescriptions and using the dosage and frequency of each drug to place prescriptions in a two-dimensional space to identify outliers in the prescription medications, medication errors are further avoided. Attached Figure Description

[0050] Figure 1This is a schematic diagram of a system for reducing medication errors by nurses according to the present invention.

[0051] Among them, 1-smartphone terminal, 2-web server, 3-pharmacy management system, 4-NFC reader Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the embodiments of this application. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0053] The present invention proposes a system for reducing nurses' medication errors, comprising: a smartphone with an application installed, a pharmacy management system, a web server, an NFC tag affixed to the medicine packaging, and an NFC reader for the pharmacy management system;

[0054] The process involves embedding a fingerprint scanner in a smartphone, which patients use to verify their identity; the smartphone application generates a timestamp, which is then sent to a web server; the application requests the active prescription ID from the web server, and uses the selected prescription ID, patient ID, and timestamp to generate patient information and send it to the web server.

[0055] The pharmacy management system receives patient information sent by patients, verifies their identity with the web server, and obtains the information required for dispensing medication.

[0056] The web server verifies the identity. If the verification is successful, the web server will send back information. The nurse scans the medicine. The medicine packaging has an NFC tag, which contains the medicine name and ID.

[0057] The pharmacy management system, after acquiring the scanned NFC tag, verifies the drug ID by matching it with the ID in the prescription information. If the match is successful, the nurse will dispense the medication.

[0058] The information required for dispensing medication includes the patient's name, medication ID, and prescription ID.

[0059] Upon first use of the application, patients are required to set up their fingerprints. Without a fingerprint, patients will not be able to access the application or transmit any information. This verification step helps verify the patient's medication dispensing process and helps protect the privacy of sensitive patient information. The active prescription ID is always stored on the web server and is not stored on the patient's mobile phone.

[0060] Thoroughly reviewing prescriptions is the final step in the medication dispensing process, extracting relevant data to prevent medication errors. The processed data contains nearly a hundred variables. This type of multivariate data is often referred to as high-dimensional data, which presents challenges for interpretation and analysis due to the curse of dimensionality, data redundancy, and noise. Any model directly built on high-dimensional data typically exhibits poor generalization performance, limiting analytical models and making it difficult to create visualizations of multidimensional data. Visualization plays a crucial role in guiding hospital management and improving nurses' medication error reporting. To improve data interpretability and achieve comprehensive visualization, appropriate variable selection, reducing data dimensionality, and identifying the optimal predictor variables are essential.

[0061] A fast parallel variable selection model is proposed to identify the best predictor of nurse-reported medication errors by using Extreme Learning Machines (ELMs) incorporated into the packaged variable selection mechanism. ELMs are accurate, extremely fast, and non-linear predictive models. During variable selection, variables are selected from the R-values ​​of each ELM model. 2 Value evaluation. The hidden layer weights of the parallelized extreme learning machine are randomly initialized and fixed without iterative adjustment. Common activation functions are applied to the hidden neurons. The only parameter learned in the parallelized extreme learning machine is the weights between the hidden and output layers. The parameters of the hidden neurons do not depend on the training data, allowing the parallelized extreme learning machine to reach an approximate optimal generalization bound.

[0062] Parallelized Extreme Learning Machines (PEMs) are used for the rapid training of single-hidden-layer feedforward networks. A single-hidden-layer feedforward neural network has three layers of neurons: the input layer provides data features and does not perform computation; the hidden layer is the only layer that undergoes a non-linear transformation; and the output layer is linear, without any transformation function or bias. The training of a parallelized extreme learning machine consists of two stages: First, random feature mapping. In this stage, the input data is randomly projected into a new space. The parallelized extreme learning machine randomly initializes the input layer weights W∈R. d*L And bias b∈R L , where L is the number of hidden neurons. Input data x∈R d The activation function is nonlinearly mapped to the hidden neuron i, as shown in Equation (1):

[0063] h i (x)=φ(w i x, b i (1)

[0064] Among them, w i ∈R d b∈R It is a non-linear piecewise continuous function. The hidden layer is not limited to having only one type of transformation function in the neurons. The number of linear neurons is equal to the number of data features, and each of these neurons replicates the corresponding feature.

[0065] The second stage of parallelized extreme learning machine is to solve the linear system and find the optimal β that minimizes the following cost function, as shown in Equation (2):

[0066]

[0067] Where H is the output of the hidden layer, as shown in formula (3):

[0068]

[0069] Where Y represents all outcome variables, Y i Let X be the outcome variable i, and X be the predictor variable for all samples. i Let S be the i-th sample of X. k Let k variables be randomly selected from X, W be the input layer weights of the parallelized extreme learning machine, b be the input layer bias of the parallelized extreme learning machine, L be the number of hidden neurons in the parallelized extreme learning machine, h be the output of the hidden neurons in the parallelized extreme learning machine, φ be the nonlinear transformation function in the parallelized extreme learning machine, β be the output layer weights of the parallelized extreme learning machine, and c be the weights of the output layer of the parallelized extreme learning machine. s Let α be the best matching unit in the self-organizing map, and σ be the learning rate in the self-organizing map. λ Let be the neighborhood function in the self-organizing map, and d be the distance function in the self-organizing map.

[0070] The cost function itself is the mean squared error between the approximation of the parallelized extreme learning machine and the true target value Y. The optimal solution of the cost function β* is shown in equation (4):

[0071] β*=H + Y (4)

[0072] Among them, H + It is a pseudo-inverse, and the calculation formula is:

[0073] H + =(H T H) -1 H T (5)

[0074] The implementation of the pseudoinverse includes a small regularization term H. + =(H T H+αI) -1 H T .

[0075] R 2The value serves as an accurate measure of the performance of variable selection. Due to the different scenarios of outcomes, the relationships between the predictor variables and each outcome variable are evaluated separately, and R0 is calculated. 2 The values ​​are divided into three prediction results for training the parallelized extreme learning machine: (X,Y1), (X,Y2), and (X,Y3).

[0076] R is calculated using the mean squared error of the parallelized extreme learning machine model. 2 The value is shown in formula (6):

[0077]

[0078] The mean squared error value is rewritten as shown in formula (7):

[0079]

[0080] In the formula, Y is the output of the parallelized extreme learning machine model. i The estimate of R. 2 The higher the value, the better the prediction. The more accurate, the better. 2 The value is used to evaluate the performance of the selected variable.

[0081] In the selection of packaging variables, multiple parallelized extreme learning machines were constructed in parallel, each evaluating k variables randomly selected from X. R0 was computed. 2 The value is used to measure the predictive performance of the selected k variables. R 2 The larger the value, the better the prediction from these selected variables, and the larger the R-value. 2 A variable whose value is [value].

[0082] The variable selection process includes the following steps:

[0083] S1: Random Variable Selection: From the original dataset X, randomly select k variables to create a subset of the original data: S k Initially, k = 1, and the current optimal variable set S k * =S k The final optimal set of variables S k ** =S k ;

[0084] S2: Parallelized Extreme Learning Machine Training. The parallelized extreme learning machine is built on the selected data S. k and outcome variable Y i Based on this, calculate R 2 value;

[0085] S3: Update the current best set of predictors. If the current R... 2If the value is higher than the previous calculation, then update S. k * =S k Otherwise, S k Keep it the same;

[0086] S4: Iterative Update: Repeat iterative steps S61 to S63 until the stopping criterion is met. The stopping criterion is defined by the maximum number of iterations. If the maximum number of iterations is reached, the final optimal variable set S is obtained. k ** By S k * renew;

[0087] S5: Increase the value of k, increase the number of variables selected by one, and start again from step S61.

[0088] Self-organizing maps (SOMs) are used to perform visualization after determining the highest predicted value. SOMs reveal the non-linear relationship between predictors and outcomes, using a predefined two-dimensional grid to capture the topological structure of high-dimensional data. Each point on the grid is assigned a weight, with each point having randomly initialized weights. After numerous iterations, these weights are updated to reflect the position of the point in the original data space. In the iterative algorithm, unit c... s For s = [1, ..., N], where N is the number of points on the two-dimensional grid, it is updated using the rule shown in Equation (8):

[0089] c s ←c s +ασ λ (r, s)(x) i -c s (8)

[0090] Where, x i Let α be the i-th data point, α be the learning rate between 0 and 1, and σ be the learning rate between 0 and 1. λ This is a neighborhood function; it returns 0 for non-neighborhoods and other non-zero values ​​for valid neighborhoods. Furthermore, d is a distance function, r = argmin{d(x...} i ,c s )}.

[0091] According to the self-organizing map algorithm, each point c on the two-dimensional grid... s s = [1,…,N] represents a set of points in a multidimensional data space. s A self-organizing map (SOM) is the optimal matching unit of a set of points in the original data space, performing discrete nonlinear dimensionality reduction. To understand the visualization, the SOM is converted into a heatmap using color, showing the importance of a given variable.

[0092] After extracting given variables, the system automatically learns thresholds between normal and abnormal dosages for each variable in e-prescriptions using previous prescriptions to detect potential outliers. Machine learning eliminates the need for extensive preprocessing, thus reducing computational costs. A density-distance centrality screening method is used; outlier detection involves identifying low-density, low-similarity prescription groups and ranking prescriptions based on their centrality indices. Overdosage or underdosage prescriptions are those with centrality scores lower than the average centrality index for each drug.

[0093] Prescription outliers are calculated by representing each prescription in a vector space, fitting each prescription drug into a two-dimensional vector using dosage and daily frequency. The relationships between prescriptions are represented as a graph, where vertices are prescriptions and edges are based on the similarity between a pair of prescriptions. A similarity function is defined as the pairwise similarity between two-dimensional vectors composed of dosage and frequency. Pairwise similarity accepts any pairwise measure; a normal prescription has a high centrality index, similar to many other prescriptions. To calculate the centrality density of each prescription, the weight of each edge is calculated using extension, weighted links, and weighted nodes to calculate the centrality score of the prescription. The weighting function is given by Equation (9):

[0094]

[0095] Among them, B u It is the set of all neighborhoods u contained in N. u Let W(v, u) represent the number of neighbors of u, and W(v, u) be the weight of the external links from v to u. The more links a prescription drug has with other prescription drugs that are highly similar to it, the more representative it is in prescription drug distribution.

[0096] This invention provides a system for reducing medication errors by nurses, making prescription information readily available to patients, providing high-quality service by minimizing medication errors, ensuring the security and privacy of patients' sensitive information, and reducing the time required to prepare prescriptions in advance. Furthermore, by combining parallelized extreme learning machines and self-organizing maps to extract relevant variables from prescriptions and using the dosage and frequency of each drug to place prescriptions in a two-dimensional space to identify outliers in the prescription medications, medication errors are further avoided.

[0097] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A system for reducing nurses' medication errors, characterized in that, include: The system includes a smartphone with an app installed, a pharmacy management system, a web server, NFC tags affixed to medicine packaging, and an NFC reader for the pharmacy management system. Among them, a fingerprint scanner is embedded in the smartphone, and the patient uses the fingerprint scanner embedded in the smartphone to verify his identity; The smartphone application generates a timestamp, and the timestamp value will be sent to the web server. The application will request a prescription ID from the web server. The application will then use the selected prescription ID, patient ID, and timestamp to generate patient information and send it to the web server. The pharmacy management system receives patient information sent by patients, verifies their identity with the web server, and obtains the information required for dispensing medication. The web server verifies the identity. If the verification is successful, the web server will send back information. The nurse scans the medicine. The medicine packaging has an NFC tag, which contains the medicine name and medicine ID. The pharmacy management system, after obtaining the scanned NFC tag, will verify the drug ID by matching it with the prescription ID. If the match is successful, the nurse will dispense the medication. To avoid medication errors, relevant data from prescriptions are extracted. A fast parallel variable selection model is proposed by using Extreme Learning Machines (ELM) integrated into the package variable selection mechanism to identify the best predictor of nurse-reported medication errors. During the variable selection process, variables are selected by the R-value of each ELM model. 2 Value assessment; After extracting given variables, the threshold between normal and abnormal doses for each variable in the electronic medical order is automatically learned using previous prescriptions to detect potential outliers in the prescription. Through machine learning, no large amount of preprocessing is required, thus reducing computational costs. The density distance centrality screening method is selected. Drug outlier detection is to find prescription groups with low density and low similarity. Prescriptions are ranked according to their centrality index. Overdose or underdose are prescriptions whose centrality scores are lower than the average centrality index of each drug. Prescription outliers are calculated by representing each prescription in a vector space, fitting each prescription drug into a two-dimensional vector using dosage and daily frequency; the relationship between prescriptions is represented as a graph, where vertices are prescriptions and edges are based on the similarity between a pair of prescriptions; a similarity function is defined, which is the pairwise similarity between two-dimensional vectors composed of dosage and frequency; pairwise similarity accepts any pairwise measure, a normal prescription has a high centrality index, it is similar to many other prescriptions; in order to calculate the centrality density of each prescription, the weight of each edge is calculated, and the centrality score of the prescription is calculated using expansion, weighted links and weighted nodes, the weighting function is shown in Equation (9): （9） Where Bu is the set of all neighborhoods u, Nv represents the number of neighborhoods of v, and W(v, u) is the weight of the external links from v to u; the more prescription drugs that are highly similar to other prescription drugs are linked together, the more representative it is in prescription drug distribution.

2. The system for reducing nurses' medication errors according to claim 1, characterized in that, The information required for dispensing medication includes the patient's name, medication ID, and prescription ID.

3. The system for reducing nurses' medication errors according to claim 2, characterized in that, The prescription ID is always stored on the web server and is not stored on the patient's mobile phone.

Images