Nursing quality evaluation and continuous improvement system and method based on machine learning

Abstract

The invention relates to the technical field of nursing quality evaluation and continuous improvement systems, in particular to a nursing quality evaluation and continuous improvement system and method based on machine learning. A patient feedback analysis module; a literature analysis module; the theme analysis module is used for analyzing the nursing quality key indexes by utilizing an LDA theme modeling technology and identifying potential problems and improvement opportunities in nursing practice; the expert knowledge base module is used for storing and providing professional knowledge support; the risk prediction module is used for predicting nursing risks based on the machine learning model; a visual interface module; a collaborative recommendation module; the combination of the federated learning core function module, the Xgboost model and the GBDT model plays a complementary role in risk prediction, the Xgboost is good at processing a nonlinear relation, the GBDT is excellent in performance when processing a linear relation, and the combination of the Xgboost and the GBDT enables the risk prediction to be more comprehensive and accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of nursing quality assessment and continuous improvement systems, and in particular to a nursing quality assessment and continuous improvement system and method based on machine learning. Background Art

[0002] With the continuous improvement of medical service quality requirements, nursing quality assessment and continuous improvement have become the top priority of medical institution management. Traditional nursing quality assessment methods mainly rely on manual observation, paper questionnaires and regular discussions. Although this method has played an important role in the past, its limitations are becoming increasingly prominent in today's information-based and digital medical environment.

[0003] First, the traditional method is highly subjective, and the evaluation results are often affected by the personal experience and preferences of the evaluator, making it difficult to ensure the objectivity and consistency of the evaluation. Secondly, this method is usually a post-evaluation, which makes it difficult to detect and warn of potential nursing risks in a timely manner. Furthermore, the traditional way of sharing best practices is inefficient, and excellent nursing experience is difficult to quickly and accurately spread to the nurses in need. In addition, with data security and patient privacy protection becoming increasingly important today, traditional methods are also difficult to achieve effective collaboration and knowledge sharing across institutions.

[0004] In the face of these challenges, the industry is constantly exploring new solutions. Some medical institutions have begun to try to introduce electronic nursing record systems and patient feedback platforms to improve the efficiency of data collection. Some researchers have also tried to apply simple statistical analysis methods to nursing quality assessment. However, although these improvements have made progress in some aspects, they have not fundamentally solved the problems of traditional methods. For example, simple electronicization cannot significantly improve the objectivity and accuracy of the assessment; basic statistical analysis also has difficulty capturing the complex patterns and potential risks in nursing practice.

[0005] More importantly, most of the existing improvement methods are isolated and lack a comprehensive solution to integrate multiple links such as data collection, analysis, risk warning, knowledge sharing and privacy protection. This leads to waste of resources and low efficiency, and also makes the continuous improvement of nursing quality face bottlenecks. Summary of the invention

[0006] In response to these problems, the present invention proposes a nursing quality assessment and continuous improvement system and method based on machine learning. By integrating multi-source data and using advanced natural language processing and machine learning technologies, the system achieves objective assessment of nursing quality, early warning of risks, intelligent recommendation of best practices, and cross-institutional collaboration under privacy protection.

[0007] The present invention proposes a nursing quality assessment and continuous improvement system based on machine learning, including:

[0008] Nursing record analysis module, used to extract key indicators of nursing quality from nursing records;

[0009] Patient feedback analysis module, used to extract key indicators of nursing quality from patient feedback;

[0010] Literature analysis module, used to extract key indicators of nursing quality from medical literature and diagnosis and treatment guidelines;

[0011] A topic analysis module, used to analyze the key indicators of nursing quality using LDA topic modeling technology to identify potential problems and improvement opportunities in nursing practice;

[0012] Expert knowledge base module, used to store and provide professional knowledge support;

[0013] Risk prediction module, used to predict nursing risks based on machine learning models;

[0014] A visualization interface module that presents real-time trends of key nursing quality indicators, abnormal detection results, and improvement suggestions;

[0015] Collaborative recommendation module, used to recommend best practices based on collaborative filtering algorithms;

[0016] Federated learning core functional module, used to achieve secure data sharing and model training across institutions;

[0017] Among them, the nursing record analysis module, the patient feedback analysis module and the literature analysis module are all connected to the topic analysis module; the topic analysis module is connected to the risk prediction module, the expert knowledge base module and the collaborative recommendation module; the risk prediction module is connected to the visualization interface module; the federated learning core function module is connected to all other modules to coordinate data interaction and information flow among modules.

[0018] Preferably, the nursing record analysis module includes:

[0019] The data acquisition unit is used to obtain nursing records from the nursing information system of the institution; the preprocessing unit is used to clean and format the nursing records; the BERT encoding unit is used to convert the preprocessed nursing records into a BERT model input sequence, and encode them through the BERT model to obtain a BERT vector; the dimension reduction unit is used to perform dimension reduction processing on the BERT vector to generate a feature vector; the indicator calculation unit is used to calculate the key indicators of nursing quality according to the feature vector, and the calculation formula is:

[0020] KI=f(PCA(BV))

[0021] Among them, KI represents the key indicator of nursing quality, f represents the nonlinear mapping function, PCA represents the principal component analysis dimensionality reduction operation, and BV represents the BERT vector.

[0022] Preferably, the structures of the patient feedback analysis module and the literature analysis module are similar to those of the nursing record analysis module, and both include a data acquisition unit, a preprocessing unit, a BERT encoding unit, a dimension reduction unit and an indicator calculation unit, wherein the data acquisition unit of the patient feedback analysis module acquires patient feedback information from the institution's information system or manual input, and the data acquisition unit of the literature analysis module acquires medical literature and treatment guidelines from the institution's digital library.

[0023] Preferably, the topic analysis module includes:

[0024] A data integration unit, used to integrate key nursing quality indicators from the nursing record analysis module, the patient feedback analysis module and the literature analysis module;

[0025] LDA modeling unit, used to perform LDA topic modeling on the integrated key indicators of nursing quality;

[0026] A weight calculation unit, used to calculate the weight between the topic and the subject word;

[0027] Clustering unit, used to cluster subject words;

[0028] The scoring unit is used to calculate the score of the topic. The calculation formula is:

[0029] S(t)=∑(w i *TF-IDF(w i ,t))

[0030] Among them, S(t) represents the score of topic t, w i represents the i-th word in topic t, TF-IDF(w i ,t) represents word w i TF-IDF value in topic t;

[0031] Visualization unit, used to visualize the LDA modeling results;

[0032] The improvement suggestion generation unit is used to generate the best improvement points and suggestions based on the topic scores and clustering results.

[0033] Preferably, the risk prediction module includes:

[0034] Xgboost model unit, which uses gradient boosting decision tree for modeling;

[0035] GBDT model unit,is modeled using a linear model;

[0036] A feature extraction unit, used to extract relevant features from nursing data;

[0037] The risk score calculation unit is used to calculate the nursing risk score based on the output of the Xgboost model and the GBDT model. The calculation formula is:

[0038] RS=α*∑(w i *X i )+β*∑(v j *Y j )

[0039] Among them, RS represents the nursing risk score, X i represents the i-th feature input of the Xgboost model, Y j represents the jth feature input of the GBDT model, w i and v j They represent the weights of the corresponding features respectively, α and β represent the overall weights of the Xgboost model and the GBDT model respectively, and α+β=1.

[0040] Preferably, the collaborative recommendation module includes:

[0041] A knowledge base construction unit, used to construct a subject heading knowledge base according to medical guidelines;

[0042] An input processing unit for processing key indicators of nursing quality from the subject analysis module;

[0043] The collaborative filtering algorithm unit is used to perform collaborative filtering calculations based on the subject word knowledge base and key indicators of nursing quality. Its core formula is:

[0044] r ui =μ+b u +b i +∑(s ij *(r uj -μ-b u -b j ))

[0045] Among them, r ui represents the predicted score of item i recommended to user u, μ represents the global average of all scores, and b u and b i They represent the bias items of user u and item i, respectively, and s ij represents the similarity between items i and j, r uj represents the actual rating of item j by user u;

[0046] The recommendation result generating unit is used to generate the best practice recommendation result based on the output of the collaborative filtering algorithm.

[0047] Preferably, the core functional modules of federated learning include: a data encryption unit, used to encrypt sensitive data of each module; a secure transmission unit, used to ensure the security of cross-institutional data transmission; a distributed training unit, used to coordinate multiple medical institutions to participate in model training without directly sharing original data; a model aggregation unit, used to integrate model updates from different institutions to generate a global model; a privacy protection unit, used to implement technologies such as differential privacy to further protect the data privacy of participating institutions.

[0048] The nursing quality assessment and continuous improvement method based on machine learning includes the following steps: (1) data acquisition and preprocessing step, including obtaining nursing records, patient feedback and medical literature, and cleaning and formatting them; (2) feature extraction step, using the BERT model to encode the preprocessed data to obtain the BERT vector, and generating the feature vector through dimensionality reduction processing; (3) key nursing quality indicator calculation step, calculating the key nursing quality indicators based on the feature vector; (4) topic analysis step, using the LDA topic model to model the key nursing quality indicators and identify potential problems and improvement opportunities; (5) risk prediction step, using Xgboost and GBDT models to predict nursing risks and calculate risk scores; (6) visualization and suggestion generation step, visualizing the topic analysis results and risk scores, and generating improvement suggestions; (7) collaborative recommendation step, recommending best practices based on collaborative filtering algorithms; (8) federated learning step, realizing secure data sharing and model training across institutions through encryption and secure transmission technology.

[0049] Preferably, the topic analysis step specifically includes: (a) integrating key indicators of nursing quality from nursing records, patient feedback and medical literature; (b) performing LDA topic modeling on the integrated indicators; (c) calculating the weights between topics and subject words; (d) clustering subject words; (e) calculating topic scores, and the scoring formula is: S(t) = Σ(w i *TF-IDF(w i ,t)); (f) visualize the modeling results; (g) generate the best improvement points and suggestions based on topic scoring and clustering results.

[0050] Preferably, the federated learning step specifically includes: (a) encrypting the sensitive data of each participating institution; (b) transmitting encrypted data between participating institutions through a secure transmission channel; (c) training the model locally without directly sharing the original data; (d) transmitting local model updates to the central server; (e) aggregating model updates from different institutions on the central server to generate a global model; (f) distributing the updated global model to each participating institution; (g) repeating steps (c) to (f) until the model converges or reaches a predetermined number of iterations.

[0051] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0052] The system of the present invention includes a nursing record analysis module, a patient feedback analysis module, a literature analysis module, a topic analysis module, an expert knowledge base module, a risk prediction module, a visualization interface module, a collaborative recommendation module, and a federated learning core function module. The organic combination of these modules not only solves complex problems that are difficult to deal with with a single technology, but also produces significant synergistic effects.

[0053] For example, the application of the BERT model in the analysis of nursing records, patient feedback, and medical literature has greatly improved the accuracy and efficiency of feature extraction. This provides high-quality input for subsequent LDA topic analysis, enabling the system to more accurately identify potential nursing problems and improvement opportunities. At the same time, the combination of Xg boost and GBDT models plays a complementary role in risk prediction. Xgboost is good at handling nonlinear relationships, while GBDT performs well in handling linear relationships. The combination of the two makes risk prediction more comprehensive and accurate.

[0054] The introduction of collaborative filtering algorithms cleverly solves the problem of inefficient dissemination of best practices. By analyzing a large amount of nursing practice data, the system can accurately recommend successful nursing methods to nurses who need them most, greatly improving the efficiency and pertinence of knowledge dissemination.

[0055] What is more worth mentioning is that the application of federated learning technology not only solves the contradiction between data sharing and privacy protection, but also opens up new ways for cross-institutional collaboration. This enables different medical institutions to jointly train more powerful and universal nursing quality assessment models while protecting patient privacy.

[0056] In general, this invention successfully transforms nursing quality assessment from a subjective, passive, and isolated mode to an objective, active, and collaborative mode through technological innovation and system integration. It not only significantly improves the accuracy and timeliness of assessment, but also provides strong support for the continuous improvement of nursing practice. This comprehensive, intelligent, and safe solution is expected to promote the entire nursing industry to develop in the direction of higher quality and higher efficiency, and ultimately bring benefits to patients, medical staff, and medical institutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is the overall logic block diagram of the system of the present invention.

[0058] Figure 2 It is a logic block diagram of the nursing record analysis module of the present invention.

[0059] Figure 3It is a logical block diagram of the subject analysis module of the present invention.

[0060] Figure 4 It is a logic block diagram of the risk prediction module of the present invention.

[0061] Figure 5 It is a logic block diagram of the collaborative recommendation module of the present invention.

[0062] Figure 6 This is a logical block diagram of the core functional module of federated learning of the present invention. DETAILED DESCRIPTION

[0063] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects thereof are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0064] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0065] Example 1

[0066] See also Figure 1-6 The present invention relates to a nursing quality assessment and continuous improvement system and method based on machine learning, aiming to improve the objectivity and comprehensiveness of nursing quality assessment and provide strong support for the continuous improvement of nursing practice.

[0067] First, the system of the present invention includes a nursing record analysis module 1, a patient feedback analysis module 2, a literature analysis module 3, a topic analysis module 4, an expert knowledge base module 5, a risk prediction module 6, a visualization interface module 7, a collaborative recommendation module 8, and a federated learning core function module 9. These modules form a complete nursing quality evaluation and improvement closed loop through carefully designed connection relationships.

[0068] Specifically, the nursing record analysis module 1, the patient feedback analysis module 2, and the literature analysis module 3 serve as the data input end of the system, respectively collecting and analyzing nursing-related information from different perspectives. These three modules are directly connected to the topic analysis module 4 to provide basic data for subsequent in-depth analysis. For example, the nursing record analysis module 1 is used to extract nurses' daily records from the electronic medical record system, the patient feedback analysis module 2 may process patient satisfaction questionnaires, and the literature analysis module 3 is used to analyze the latest nursing guidelines and research literature.

[0069] The topic analysis module 4 is then connected to the risk prediction module 6, the expert knowledge base module 5 and the collaborative recommendation module 8 to form the core analysis and decision-making part of the system. This design allows the system to not only identify current nursing problems, but also predict potential risks and provide improvement suggestions based on expert knowledge and best practices. For example, if the topic analysis module 4 finds that the theme of "inadequate patient pain management" appears repeatedly in multiple nursing records, the risk prediction module 6 is used to warn that this may lead to the risk of decreased patient satisfaction and delayed recovery, while the collaborative recommendation module 8 is used to recommend some successful pain management solutions.

[0070] Finally, the risk prediction module 6 is connected to the visualization interface module 7 to present the analysis results to the user in an intuitive manner. This design enables nursing managers to quickly understand the current nursing quality status and potential risks, so as to make timely and accurate decisions.

[0071] Preferably, the federated learning core function module 9 is connected to all other modules. This design ensures that data sharing and model training across institutions are achieved while protecting data privacy. For example, multiple hospitals can jointly train a more accurate nursing risk prediction model without directly sharing sensitive patient information.

[0072] Next, the nursing record analysis module 1 includes a data acquisition unit 11, a preprocessing unit 12, a BERT encoding unit 13, a dimension reduction unit 14 and an index calculation unit 15. This design enables the system to extract valuable key indicators of nursing quality from the original nursing records.

[0073] Specifically, the data acquisition unit 11 first acquires nursing records from the hospital's nursing information system. These records may include nurses' daily observation records, nursing plans, nursing operation records, etc. Then, the preprocessing unit 12 cleans and formats these records, such as removing irrelevant information, unifying date formats, etc.

[0074] Next, the BERT encoding unit 13 converts the preprocessed nursing records into an input sequence for the BERT model. BERT (Bidirectional Encoder Representations from Transformers) is an advanced natural language processing model that can understand the contextual semantics of text. Through the BERT model, the system can convert unstructured nursing records into structured vector representations.

[0075] Afterwards, the dimension reduction unit 14 performs dimension reduction processing on the BERT vector to generate a feature vector. This step is intended to reduce the dimension of the data and improve the efficiency of subsequent processing. Finally, the indicator calculation unit 15 calculates the key indicators of nursing quality based on the feature vector.

[0076] In one embodiment of the present invention, the calculation formula of the key indicator of nursing quality is: KI = f(PCA(BV)). Wherein, KI represents the key indicator of nursing quality, f represents the nonlinear mapping function, PCA represents the principal component analysis dimensionality reduction operation, and BV represents the BERT vector. This design allows the system to extract key quality indicators from complex nursing records, such as nursing plan completion rate, patient comfort score, etc.

[0077] Preferably, the number of principal components retained by the PCA dimensionality reduction operation can be set to 20% to 30% of the original BERT vector dimension, which can significantly reduce the computational complexity while retaining most of the information. For example, if the original BERT vector dimension is 768, the dimension of the feature vector after dimensionality reduction can be set to between 150 and 230.

[0078] Again, the structures of the patient feedback analysis module 2 and the literature analysis module 3 are similar to those of the nursing record analysis module 1, and also include a data acquisition unit, a preprocessing unit, a BERT encoding unit, a dimension reduction unit, and an indicator calculation unit. This consistent design enables the system to process data from different sources in a similar way, thereby improving the overall efficiency and consistency of the system.

[0079] For example, the data acquisition unit of the patient feedback analysis module 2 may obtain data from the hospital's patient satisfaction survey system, while the data acquisition unit of the literature analysis module 3 may obtain data from the hospital's digital library or a public medical literature database. Although the data sources are different, the subsequent processing procedures are basically the same, which greatly simplifies the design and maintenance of the system.

[0080] Then, the topic analysis module 4 includes a data integration unit 41, an LDA modeling unit 42, a weight calculation unit 43, a clustering unit 44, a scoring unit 45, a visualization unit 46 and an improvement suggestion generation unit 47. This design enables the system to deeply analyze nursing quality issues from multiple perspectives and generate valuable improvement suggestions.

[0081] Specifically, the data integration unit 41 first integrates the key indicators of nursing quality from the nursing record analysis module 1, the patient feedback analysis module 2 and the literature analysis module 3. This integration of multi-source data provides a more comprehensive perspective for subsequent analysis. Then, the LDA modeling unit 42 performs LDA (Latent Dirichlet Allocation) topic modeling on the integrated indicators. LDA is a commonly used text topic modeling technology that can discover potential topics from a large amount of text.

[0082] Next, the weight calculation unit 43 calculates the weights between the topics and the subject words, and the clustering unit 44 clusters the subject words. These two steps help the system better understand and organize the discovered topics. The scoring unit 45 then calculates the score of the topic, and its calculation formula is:

[0083] S(t)=Σ(w i *TF-IDF(w i ,t))

[0084] Among them, S(t) represents the score of topic t, w i represents the i-th word in topic t, TF-IDF(w i ,t) represents word w i TF-IDF value in topic t.

[0085] Preferably, when calculating the TF-IDF value, a logarithmic smoothed IDF formula may be used:

[0086]

[0087] Where N is the total number of documents and DF(w) is the number of documents containing word w. This method can avoid the problem of too large IDF values ​​for rare words.

[0088] Finally, the visualization unit 46 visualizes the LDA modeling results, and the improvement suggestion generation unit 47 generates the best improvement points and suggestions based on the topic scores and clustering results. For example, if the system finds that the score of the topic "patient communication" is low and is clustered with the subject words "nurse-patient relationship" and "information communication", the system is used to generate an improvement suggestion of "strengthening nurse-patient communication skills training".

[0089] Secondly, the risk prediction module 6 includes an Xgboost model unit 61, a GBDT model unit 62, a feature extraction unit 63, and a risk score calculation unit 64. This design combines two powerful machine learning models to more accurately predict nursing risks.

[0090] Specifically, the Xgboost model unit 61 uses a gradient boosting decision tree for modeling, while the GBDT model unit 62 uses a linear model for modeling. The two models have their own characteristics. Xgboost is good at processing nonlinear relationships, while GBDT performs well in processing linear relationships. The feature extraction unit 63 extracts relevant features from the nursing data to provide input for the two models.

[0091] The full name of Xgboost model unit 61 is eXtreme Gradient Boosting, which stands for extreme gradient boosting. Xgboost is an ensemble learning algorithm based on decision trees. It optimizes the model by iteratively adding new decision trees, and each iteration attempts to reduce the residual of the previous model. The full name of GBDT model unit 62 is GradientBoosting Decision Tree, which stands for gradient boosting decision tree. GBDT is an ensemble learning method that uses decision trees as base learners and minimizes the loss function by iteratively training decision trees.

[0092] In the present invention, the algorithm formula of the Xgboost model unit 61 is:

[0093]

[0094] Among them, y pred is the predicted value, f k is the kth decision tree, x i are input features.

[0095] Objective function:

[0096]

[0097] Among them, l is the loss function and Ω is the regularization term.

[0098] The algorithm formula of GBDT model unit 62 is:

[0099] F m (x) = F m-1 (x)+γ m ·h m (x)

[0100] Among them, F m (x) is the model after the mth iteration, h m (x) is the newly added decision tree, γ m is the step length.

[0101] In the present invention, these two models work together to more accurately predict nursing risks. Specifically:

[0102] The feature extraction unit 63 first extracts relevant features from the nursing data, such as the patient's age, medical history, current medication, vital signs, etc. The Xgboost model unit 61 mainly processes nonlinear feature relationships. For example, it is used to discover the nonlinear relationship between age and certain disease risks, or the impact of complex interactions between multiple drugs on nursing risks. The GBDT model unit 62 focuses on processing linear feature relationships. For example, it is used to better capture the linear relationship between vital signs such as blood pressure and heart rate and certain nursing risks.

[0103] Finally, the risk score calculation unit 64 calculates the nursing risk score according to the output of the Xgboost model and the GBDT model, and the calculation formula is: RS = α * ∑ (w i *X i )+β*∑(v j *Y j )

[0104] Among them, RS represents the nursing risk score, X i represents the i-th feature input of the Xgboost model, Y j represents the jth feature input of the GBDT model, w i and v j They represent the weights of the corresponding features respectively, α and β represent the overall weights of the Xgboost model and the GBDT model respectively, and α+β=1.

[0105] In one embodiment of the present invention, α=0.6 and β=0.4 can be set, which fully utilizes the advantages of the Xgboost model in processing nonlinear relationships and considers the contribution of the GBDT model. At the same time, these two parameters can be dynamically adjusted according to the needs of the actual application scenario.

[0106] In this way, the system can take into account both linear and nonlinear factors in nursing risks and improve the accuracy of predictions. For example, when predicting a patient's fall risk, Xgboost is used to capture the complex interactions between age, medication type and quantity, while GBDT is used to better handle the linear relationship between gait scores, muscle strength, etc. and fall risk. The combination of the two can provide a more comprehensive and accurate risk assessment.

[0107] Then, the collaborative recommendation module 8 includes a knowledge base construction unit 81, an input processing unit 82, a collaborative filtering algorithm unit 83 and a recommendation result generation unit 84. This design enables the system to recommend the best nursing practice based on historical data and current conditions.

[0108] Specifically, the knowledge base construction unit 81 first constructs a subject word knowledge base based on the medical guidelines. This knowledge base contains standard nursing practices and related terms. The input processing unit 82 processes the key indicators of nursing quality from the subject analysis module 4. The collaborative filtering algorithm unit 83 performs collaborative filtering calculations based on the subject word knowledge base and the key indicators of nursing quality, and its core formula is:

[0109] r ui =μ+b u +b i +Σ(s ij *(r uj -μ-b u -b j ))

[0110] Among them, r ui represents the predicted score of item i recommended to user u, μ represents the global average of all scores, and b u and b i They represent the bias items of user u and item i, respectively, and s ij represents the similarity between items i and j, r uj represents the actual rating of user u on item j.

[0111] Preferably, when calculating the item similarity s ij When , cosine similarity can be used, and its formula is:

[0112]

[0113] This method can effectively capture the similarities between items and is not affected by the rating scale.

[0114] Finally, the recommendation result generation unit 84 generates the best practice recommendation result based on the output of the collaborative filtering algorithm. For example, if the system finds that one nursing team has a low score in pain management, while the other team performs well in this regard, the system is used to recommend the latter's pain management practices to the former.

[0115] Thirdly, the federated learning core functional module 9 includes a data encryption unit 91, a secure transmission unit 92, a distributed training unit 93, a model aggregation unit 94, and a privacy protection unit 95. This design ensures that while cross-institutional data sharing and model training are achieved, the privacy of patients is fully protected.

[0116] Specifically, the data encryption unit 91 first encrypts the sensitive data of each module. Then, the secure transmission unit 92 ensures the security of cross-institutional data transmission. The distributed training unit 93 coordinates multiple medical institutions to participate in model training without directly sharing the original data. The model aggregation unit 94 integrates model updates from different institutions to generate a global model. Finally, the privacy protection unit 95 implements technologies such as differential privacy to further protect the data privacy of participating institutions.

[0117] In one embodiment of the present invention, homomorphic encryption technology can be used to encrypt data. This technology allows calculations to be performed on encrypted data without decryption, thereby achieving effective use of data while protecting privacy. At the same time, the privacy budget ε of differential privacy can be set to 0.1, which has been proven in practice to achieve a good balance between protecting privacy and maintaining data usefulness.

[0118] Finally, the present invention also provides a method for nursing quality assessment and continuous improvement based on machine learning. This method includes steps such as data acquisition and preprocessing, feature extraction, calculation of key indicators of nursing quality, topic analysis, risk prediction, visualization and suggestion generation, collaborative recommendation and federated learning.

[0119] Specifically, in the data acquisition and preprocessing step, the system first obtains nursing records, patient feedback, and medical literature from various sources, and cleans and formats them. For example, it may be necessary to remove typos in nursing records, unify date formats, etc. Then in the feature extraction step, the system uses the BERT model to encode the preprocessed data, obtain the BERT vector, and generate a feature vector through dimensionality reduction. For example, for a nursing record "The patient's pain score today is 3, which is an improvement from yesterday", the BERT model is used to generate a 768-dimensional vector, which is then reduced to a feature vector of about 200 dimensions through PCA.

[0120] Next, in the step of calculating the key indicators of nursing quality, the system calculates the key indicators of nursing quality based on the feature vectors. These indicators may include pain management effect, patient satisfaction, nursing plan execution rate, etc. For example, the system is used to calculate a pain management effect score between 0 and 100 based on the feature vectors of multiple nursing records.

[0121] Then, in the topic analysis step, the system uses the LDA topic model to model key indicators of nursing quality and identify potential problems and improvement opportunities. This step specifically includes sub-steps such as integrating indicators, LDA modeling, calculating weights, clustering, calculating scores, visual processing, and generating improvement suggestions. For example, the system is used to find that "pain management", "patient communication" and "nursing documentation" are the three main topics, among which "pain management" has the lowest score, so it needs to be improved.

[0122] In the risk prediction step, the system uses Xgboost and GBDT models to predict nursing risks and calculate risk scores. For example, based on the patient’s age, disease type, nursing records and other characteristics, the system predicts that the patient has a 30% risk of falling in the next 24 hours.

[0123] Next, in the visualization and suggestion generation step, the system visualizes the topic analysis results and risk scores and generates improvement suggestions. For example, the system is used to generate a dashboard that displays the scores and changing trends of various nursing quality indicators, and marks a red warning in the "pain management" section, while providing a suggestion to "strengthen nurse pain assessment training."

[0124] In the collaborative recommendation step, the system recommends best practices based on collaborative filtering algorithms. For example, if the system finds that nurse A performs well in pain management, but nurse B needs to improve in this regard, the system is used to recommend nurse A's pain management method to nurse B.

[0125] In the federated learning step, the system uses encryption and secure transmission technology to achieve secure data sharing and model training across institutions. This step specifically includes sub-steps such as data encryption, secure transmission, local training, model update transmission, global model aggregation and distribution. Specifically, it includes: (a) encrypting the sensitive data of each participating institution; (b) transmitting encrypted data between participating institutions through a secure transmission channel; (c) training the model locally without directly sharing the original data; (d) transmitting local model updates to the central server; (e) aggregating model updates from different institutions on the central server to generate a global model; (f) distributing the updated global model to each participating institution; (g) repeating steps (c) to (f) until the model converges or reaches a predetermined number of iterations. For example, multiple hospitals are used to jointly train a nursing risk prediction model. Each hospital trains the model locally, and then only transmits the model updates (not the original data) to the central server for aggregation.

[0126] In one embodiment of the present invention, the number of iterations of federated learning can be set to 100, and the performance index (such as AUC value) of the model is calculated after each iteration. If the model performance does not improve significantly after 5 consecutive iterations (such as the AUC value improvement is less than 0.001), the training process can be terminated in advance. In this way, unnecessary waste of computing resources can be avoided while ensuring the performance of the model.

[0127] Preferably, when performing model aggregation, a weighted average method can be used, with the weights proportional to the amount of data from each participating institution. This ensures that institutions with more data have a greater influence on the final model, thereby improving the overall performance of the model.

[0128] Through the above steps, the method of the present invention can comprehensively evaluate the quality of care, timely predict potential risks, and provide targeted improvement suggestions. At the same time, through federated learning technology, the method can also achieve cross-institutional knowledge sharing and model optimization under the premise of protecting patient privacy.

[0129] In general, the system and method of the present invention integrates multi-source data and uses advanced machine learning technology to achieve objective evaluation of nursing quality, early warning of risks, recommendation of best practices, and cross-institutional collaboration under privacy protection. This not only helps to improve nursing quality and reduce medical risks, but also promotes the dissemination of nursing knowledge and continuous improvement of nursing practice, ultimately benefiting patients, medical staff, and medical institutions.

[0130] In order to verify the superiority of the present invention, we conducted a comparative experiment to compare the embodiments of the present invention with the traditional nursing quality assessment method. The following is a detailed description of the embodiments and comparative examples and an analysis of the test results.

[0131] Example 1: In the medical ward of a tertiary hospital, we deployed the nursing quality assessment and continuous improvement system based on machine learning of the present invention. The system integrates nursing records, patient feedback and the latest medical literature, extracts features through the BERT model, identifies potential problems using the LDA topic model, and combines the Xgboost and GBDT models for risk prediction. At the same time, the system also recommends best practices through collaborative filtering algorithms and uses federated learning technology to achieve secure data sharing with two other partner hospitals.

[0132] Comparative Example 1: In the surgical ward of the same hospital, the nursing department continued to use the traditional nursing quality assessment method. This method mainly relies on regular inspections by the head nurse, paper patient satisfaction questionnaires, and monthly nursing quality discussions. Risk warnings are mainly based on the experience of nurses, and sharing of best practices is carried out through regular in-hospital training.

[0133] We conducted a six-month comparative test of the two methods, focusing on the following indicators: nursing quality score accuracy, risk warning timeliness, best practice adoption rate, patient satisfaction, and nurse work efficiency. The detection methods and standards for these indicators are as follows:

[0134] 1. Accuracy of nursing quality scores: An independent team of experts scored 100 randomly selected nursing records, compared them with the system / manual scores, and calculated the accuracy.

[0135] 2. Timely rate of risk warning: The proportion of cases with successful warning and timely intervention in the total risk events within 6 months is counted.

[0136] 3. Best Practice Adoption Rate: This is the percentage of best practices that are actually applied in daily work as recommended by the tracking system / taught in training sessions.

[0137] 4. Patient satisfaction: The satisfaction score is calculated through a standardized patient satisfaction questionnaire.

[0138] 5. Nurse work efficiency: Statistics on the time nurses spend on quality assessment and improvement related work.

[0139] The test results are shown in the following table:

[0140]

[0141]

[0142] It can be seen from the test results that the embodiments of the present invention are significantly superior to traditional methods in all indicators. It is particularly noteworthy that the accuracy of nursing quality scoring has increased by 14 percentage points, which shows that the evaluation method based on machine learning can more objectively and comprehensively reflect the actual nursing quality level. The improvement in the timeliness of risk warning has reached 23 percentage points, which means that this system can effectively reduce medical risks and improve patient safety.

[0143] The significant increase in the adoption rate of best practices (23 percentage points) shows that the recommendation algorithm based on collaborative filtering can more accurately promote effective nursing methods to areas that need improvement. This not only promotes the dissemination of nursing knowledge, but also directly improves the overall quality of nursing, which can be confirmed by the improvement in patient satisfaction (8 points).

[0144] Another important finding is the 30% improvement in nurses’ productivity, which indicates that the system not only improves the quality of care, but also reduces the workload of nurses, allowing them to devote more time and energy to direct patient care.

[0145] It is worth mentioning that the federated learning technology used in Example 1 enables the hospital to benefit from the experience of the other two hospitals without directly sharing sensitive data. This not only expands the data sample and improves the accuracy of the model, but also provides new possibilities for cross-hospital collaboration.

[0146] In summary, the nursing quality assessment and continuous improvement system based on machine learning of the present invention has shown significant advantages in improving assessment accuracy, strengthening risk warning, promoting the dissemination of best practices, improving patient satisfaction and optimizing nurse work efficiency. This data-driven intelligent method can not only better support the daily nursing work of medical institutions, but also provide a powerful tool for the continuous improvement of nursing quality. In the future, with the accumulation of more data and further optimization of algorithms, we have reason to believe that this system will make a greater contribution to improving the overall quality of medical services.

[0147] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. Nursing quality assessment and continuous improvement system based on machine learning, characterized by: include: Nursing record analysis module, used to extract key indicators of nursing quality from nursing records; Patient feedback analysis module, used to extract key indicators of nursing quality from patient feedback; Literature analysis module, used to extract key indicators of nursing quality from medical literature and diagnosis and treatment guidelines; A topic analysis module, used to analyze the key indicators of nursing quality using LDA topic modeling technology to identify potential problems and improvement opportunities in nursing practice; Expert knowledge base module, used to store and provide professional knowledge support; Risk prediction module, used to predict nursing risks based on machine learning models; A visualization interface module that presents real-time trends of key nursing quality indicators, abnormal detection results, and improvement suggestions; Collaborative recommendation module, used to recommend best practices based on collaborative filtering algorithms; Federated learning core functional module, used to achieve secure data sharing and model training across institutions; Among them, the nursing record analysis module, the patient feedback analysis module and the literature analysis module are all connected to the topic analysis module; the topic analysis module is connected to the risk prediction module, the expert knowledge base module and the collaborative recommendation module; the risk prediction module is connected to the visualization interface module; the federated learning core function module is connected to all other modules to coordinate data interaction and information flow among modules.

2. The system according to claim 1, characterized in that The nursing record analysis module includes: The data acquisition unit is used to obtain nursing records from the nursing information system of the institution; the preprocessing unit is used to clean and format the nursing records; the BERT encoding unit is used to convert the preprocessed nursing records into a BERT model input sequence, and encode them through the BERT model to obtain a BERT vector; the dimension reduction unit is used to perform dimension reduction processing on the BERT vector to generate a feature vector; the indicator calculation unit is used to calculate the key indicators of nursing quality according to the feature vector, and the calculation formula is: KI=f(PCA(BV)) Among them, KI represents the key indicator of nursing quality, f represents the nonlinear mapping function, PCA represents the principal component analysis dimensionality reduction operation, and BV represents the BERT vector.

3. The system according to claim 1, characterized in that The structures of the patient feedback analysis module and the literature analysis module are similar to those of the nursing record analysis module, and both include a data acquisition unit, a preprocessing unit, a BERT encoding unit, a dimension reduction unit, and an indicator calculation unit. The data acquisition unit of the patient feedback analysis module acquires patient feedback information from the institution's information system or manual input, and the data acquisition unit of the literature analysis module acquires medical literature and treatment guidelines from the institution's digital library.

4. The system according to claim 1, characterized in that The topic analysis module includes: A data integration unit, used to integrate key nursing quality indicators from the nursing record analysis module, the patient feedback analysis module and the literature analysis module; LDA modeling unit, used to perform LDA topic modeling on the integrated key indicators of nursing quality; A weight calculation unit, used to calculate the weight between the topic and the subject word; Clustering unit, used to cluster subject words; The scoring unit is used to calculate the score of the topic. The calculation formula is: S(t)=Σ(w i *TF-IDF(w i ,t)) Among them, S(t) represents the score of topic t, w i represents the i-th word in topic t, TF-IDF(w i ,t) represents word w i TF-IDF value in topic t; Visualization unit, used to visualize the LDA modeling results; The improvement suggestion generation unit is used to generate the best improvement points and suggestions based on the topic scores and clustering results.

5. The system according to claim 1, characterized in that The risk prediction module comprises: Xgboost model unit, which uses gradient boosting decision tree for modeling; GBDT model unit,is modeled using a linear model; A feature extraction unit, used to extract relevant features from nursing data; The risk score calculation unit is used to calculate the nursing risk score based on the output of the Xgboost model and the GBDT model. The calculation formula is: RS=α*∑(w i *X i )+β*∑(v j *Y j ) Among them, RS represents the nursing risk score, X i represents the i-th feature input of the Xgboost model, Y j represents the jth feature input of the GBDT model, w i and v j They represent the weights of the corresponding features respectively, α and β represent the overall weights of the Xgboost model and the GBDT model respectively, and α+β=1.

6. The system according to claim 1, characterized in that The collaborative recommendation module includes: A knowledge base construction unit, used to construct a subject heading knowledge base according to medical guidelines; An input processing unit for processing key indicators of nursing quality from the subject analysis module; The collaborative filtering algorithm unit is used to perform collaborative filtering calculations based on the subject word knowledge base and key indicators of nursing quality. Its core formula is: r ui =μ+b u +b i +Σ(s ij *(r uj -μ-b u -b j )) Among them, r ui represents the predicted score of item i recommended to user u, μ represents the global average of all scores, and b u and b i They represent the bias items of user u and item i, respectively, and s ij represents the similarity between items i and j, r uj represents the actual rating of item j by user u; The recommendation result generating unit is used to generate the best practice recommendation result based on the output of the collaborative filtering algorithm.

7. The system according to claim 1, characterized in that The core functional modules of federated learning include: a data encryption unit, which is used to encrypt the sensitive data of each module; a secure transmission unit, which is used to ensure the security of cross-institutional data transmission; a distributed training unit, which is used to coordinate multiple medical institutions to participate in model training without directly sharing the original data; a model aggregation unit, which is used to integrate model updates from different institutions to generate a global model; a privacy protection unit, which is used to implement technologies such as differential privacy to further protect the data privacy of participating institutions.

8. The method for evaluating and continuously improving nursing quality based on machine learning according to any one of claims 1 to 7, characterized in that: The following steps are involved: (1) Data acquisition and preprocessing steps, including obtaining nursing records, patient feedback, and medical literature, and cleaning and formatting them; (2) Feature extraction step, using the BERT model to encode the preprocessed data to obtain the BERT vector, and then generating the feature vector through dimensionality reduction processing; (3) Nursing quality key indicator calculation step, calculating the nursing quality key indicators based on the feature vector; (4) Topic analysis step, using the LDA topic model to model the nursing quality key indicators and identify potential problems and improvement opportunities; (5) Risk prediction step, using the Xgboost and GBDT models to predict nursing risks and calculate risk scores; (6) Visualization and suggestion generation step, which visualizes the topic analysis results and risk scores and generates improvement suggestions; (7) Collaborative recommendation step, which recommends best practices based on collaborative filtering algorithms; (8) Federated learning step, which achieves secure data sharing and model training across institutions through encryption and secure transmission technology.

9. The method according to claim 8, characterized in that The topic analysis steps specifically include: (a) integrating key indicators of nursing quality from nursing records, patient feedback and medical literature; (b) performing LDA topic modeling on the integrated indicators; (c) calculating the weights between topics and subject words; (d) clustering subject words; (e) calculating topic scores, and the scoring formula is: S(t) = Σ(w i *TF-IDF(w i ,t)); (f) visualize the modeling results; (g) generate the best improvement points and suggestions based on topic scoring and clustering results.

10. The method according to claim 8, characterized in that The federated learning steps specifically include: (a) encrypting the sensitive data of each participating institution; (b) transmitting encrypted data between participating institutions through a secure transmission channel; (c) training the model locally without directly sharing the original data; (d) transmitting local model updates to the central server; (e) aggregating model updates from different institutions on the central server to generate a global model; (f) distributing the updated global model to each participating institution; (g) repeating steps (c) to (f) until the model converges or reaches a predetermined number of iterations.

Images