An infectious tophi ulcer wound care regimen evaluation method and system

Abstract

The present application relates to a kind of infectious gouty stone ulcer wound care program evaluation method and system, the standardized electronic medical record data of patient is collected through medical information system, including demographic characteristics, wound characteristics, laboratory test index sequence and nursing record;After the data are structured, time stamp is unified and missing value is filled, input into pre-trained deep learning model, the expected wound healing probability, the predicted healing time and complication risk score corresponding to different nursing scheme are output, and then a comprehensive evaluation report is generated, the report is stored in database and visual query interface is provided, while collecting actual nursing result data for continuous optimization update of model.The present application realizes the quantitative evaluation and prediction of nursing scheme, effectively assists clinical decision-making, and improves the precision and resource utilization efficiency of nursing management.

Description

Technical Field

[0001] This application relates to the field of medical big data technology, and in particular to a method and system for evaluating the nursing care plan for infectious gouty tophi ulcer wounds. Background Technology

[0002] Infected tophi are a serious complication of late-stage gout, and their wound care is highly complex, requiring comprehensive consideration of individual patient differences, ulcer characteristics, and dynamic changes in the condition. Traditional nursing care assessments rely primarily on clinical experience, lacking quantitative decision support tools and struggling to systematically integrate multi-source medical data to accurately predict the effects of different care plans. This experience-driven approach is highly subjective and fails to fully utilize the patterns implicit in historical case data.

[0003] While existing electronic medical record systems can record some nursing-related information, the data is often scattered across different modules and formatted inconsistently, lacking a structured data processing method specifically for gouty tophi ulcers. Wound characteristics described in natural language are difficult to use directly for analysis, and time-series information from laboratory indicators has not been effectively utilized. Furthermore, conventional statistical methods struggle to handle high-dimensional, nonlinear clinical data, making it impossible to establish accurate predictive models to assess the expected effects and resource consumption of nursing protocols.

[0004] Therefore, an intelligent assessment solution is needed that can integrate multi-dimensional medical data and possess continuous learning capabilities. By structurally processing heterogeneous clinical data and utilizing deep learning technology to uncover potential patterns in historical data, quantitative predictions of the healing effects, time cycles, and safety of different nursing protocols can be achieved. This systematic assessment method not only provides objective evidence for clinical decision-making but also continuously optimizes prediction accuracy through ongoing data accumulation, ultimately improving nursing quality and resource utilization efficiency. Summary of the Invention

[0005] To achieve the above objectives, this application provides the following technical solution:

[0006] According to a first aspect of the present invention, the present invention claims protection for a method for evaluating the care plan of infected gouty tophi ulcer wounds, comprising:

[0007] Step S101: Collect standardized electronic medical record data of patients with gouty tophi ulcers through the medical information system. The standardized electronic medical record data includes demographic information, ulcer wound characteristic records, laboratory test index sequences, and previous nursing operation records.

[0008] Step S102: The collected standardized electronic medical record data is structured, the timestamp format is unified, missing numerical laboratory indicators are filled using time series-based interpolation, and categorical variables are filled using mode filling.

[0009] Step S103: Input the preprocessed standardized electronic medical record data into the pre-trained nursing effect prediction model, and use a deep neural network architecture to output the expected wound healing probability and expected healing time corresponding to different nursing plans.

[0010] Step S104: Analyze the output of the prediction model and generate a nursing plan evaluation report based on the expected wound healing probability, expected healing time, and medical resource consumption parameters;

[0011] Step S105: Store the nursing plan evaluation report in the medical decision support database and provide a visual query interface for clinical medical staff;

[0012] Step S106: Collect the actual nursing plan implemented and its corresponding wound healing results data, and update it to the medical decision support database.

[0013] Furthermore, step S102 also includes:

[0014] During structured processing, natural language processing is performed on the ulcer wound feature records to extract key medical entities and convert them into standardized medical terminology codes.

[0015] Furthermore, step S103 also includes:

[0016] When constructing the pre-trained nursing effect prediction model:

[0017] Step S301: Collect historical medical record data of patients with tophi ulcers from the medical data warehouses of multiple medical institutions, including basic patient characteristics, ulcer severity scores, inflammatory marker test results, microbial culture results, detailed records of implemented nursing plans, and objective assessment results of wound healing;

[0018] Step S302: Clean and standardize the collected historical gouty tophi ulcer patient medical records, unify ulcer staging standards, laboratory indicator units and nursing operation terminology from different sources, and establish a cross-institutional standard data model;

[0019] Step S303: Construct a multi-task deep learning network architecture, including a feature extraction shared layer and multiple task-specific output layers, to simultaneously predict wound healing probability, healing time, and risk of complications;

[0020] Step S304: Divide the standardized historical gouty tophi ulcer patient medical records into training set, validation set and test set in chronological order, optimize network parameters using a batch training strategy, and use a weighted loss function to balance the importance of different tasks during training.

[0021] Step S305: Monitor model performance on the validation set, use an early stopping strategy to prevent overfitting, and finally evaluate the model's prediction accuracy and generalization ability on the test set.

[0022] Step S306: Deploy the performance-compliant model to the production environment and encapsulate it into an application programming interface (API) that can provide real-time prediction services.

[0023] Furthermore, step S106 includes model optimization and updating:

[0024] Step S401: Periodically extract newly accumulated nursing plan execution data and wound healing outcome data from the medical decision support database;

[0025] Step S402: Perform quality control and standardization on the newly extracted data to ensure that it is consistent with the training data format, and label the data with timestamps;

[0026] Step S403: Employ an online learning algorithm to incrementally train the deployed nursing effect prediction model using new data and dynamically adjust the model parameters;

[0027] Step S404: Detect the prediction effect of the model after incremental training through the model performance monitoring system, and deploy a new version of the model when the performance improvement exceeds the set threshold;

[0028] Step S405: Record the metadata for each model update, including update time, data volume, performance change metrics, and store it in the model version management database.

[0029] Furthermore, the medical resource consumption parameters mentioned in step S104 include nursing hours, dressing consumption, drug costs, and equipment usage costs.

[0030] Furthermore, the multi-task deep learning network architecture described in step S303 also includes an attention mechanism module to automatically identify clinical feature variables that are important to the prediction results.

[0031] Furthermore, the visual query interface described in step S105 provides multi-dimensional data analysis functions, including filtering and comparing the effectiveness evaluation results of different nursing plans based on conditions such as ulcer severity, patient age group, and comorbidity type.

[0032] Furthermore, step S301 also includes a process for de-identifying sensitive patient information, where all personally identifiable information is replaced with anonymous identifiers before model training.

[0033] Furthermore, the nursing effect prediction model described in step S103 also outputs a complication risk score, and in step S104, the safety of the nursing plan is assessed based on the complication risk score.

[0034] The stored procedure in step S105 uses a distributed database system and ensures patient privacy and security through data encryption and access control mechanisms.

[0035] According to a second aspect of the present invention, the present invention claims protection for an assessment system for the care of infected gouty tophi ulcer wounds, comprising:

[0036] A memory having stored one or more programs that, when executed by one or more processors, cause the one or more processors to implement the aforementioned method for evaluating the care of infected gouty tophi ulcer wounds.

[0037] This invention relates to a method and system for evaluating nursing care plans for infected gouty tophi ulcer wounds. The method involves collecting standardized electronic medical record data from patients through a medical information system, including demographics, wound characteristics, laboratory test result sequences, and nursing records. After structured processing, timestamp unification, and missing value imputation, the data is input into a pre-trained deep learning model. The model outputs the expected wound healing probability, estimated healing time, and complication risk score corresponding to different nursing plans, thereby generating a comprehensive evaluation report. This report is stored in a database and provides a visual query interface. Simultaneously, actual nursing outcome data is collected for continuous model optimization and updates. This invention achieves quantitative evaluation and prediction of nursing plans, effectively assisting clinical decision-making and improving the accuracy of nursing management and resource utilization efficiency. Attached Figure Description

[0038] Figure 1 A flowchart illustrating the evaluation method for the care of infected tophi ulcer wounds as claimed in this application;

[0039] Figure 2 A second workflow diagram of an assessment method for the care of an infected tophi ulcer wound, as claimed in an embodiment of this application;

[0040] Figure 3 This is a third workflow diagram of an assessment method for the care of an infected tophi ulcer wound, as claimed in an embodiment of this application. Detailed Implementation

[0041] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0042] The terms "first," "second," and "third" in this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationships and movements between components in a specific orientation (as shown in the figures). If the specific orientation changes, the directional indications also change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0043] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0044] According to a first embodiment of the present invention, the present invention claims protection for an assessment method for the care of infected gouty tophi ulcer wounds, referring to... Figure 1 ,include:

[0045] Step S101: Collect standardized electronic medical record data of patients with gouty tophi ulcers through the medical information system. The standardized electronic medical record data includes demographic information, ulcer wound characteristic records, laboratory test index sequences, and previous nursing operation records.

[0046] Step S102: The collected standardized electronic medical record data is structured, the timestamp format is unified, missing numerical laboratory indicators are filled using time series-based interpolation, and categorical variables are filled using mode filling.

[0047] Step S103: Input the preprocessed standardized electronic medical record data into the pre-trained nursing effect prediction model, and use a deep neural network architecture to output the expected wound healing probability and expected healing time corresponding to different nursing plans.

[0048] Step S104: Analyze the output of the prediction model and generate a nursing plan evaluation report based on the expected wound healing probability, expected healing time, and medical resource consumption parameters;

[0049] Step S105: Store the nursing plan evaluation report in the medical decision support database and provide a visual query interface for clinical medical staff;

[0050] Step S106: Collect the actual nursing plan implemented and its corresponding wound healing results data, and update it to the medical decision support database.

[0051] In this embodiment, step S101 involves collecting complete electronic medical records of patients diagnosed with infectious tophi ulcers daily through the hospital information system interface. The records include basic information such as the patient's age, gender, and duration of illness; a detailed description of the location, size, depth, nature of exudate, and condition of the surrounding skin of the ulcer; a time series of laboratory test values ​​such as serum uric acid, white blood cell count, and C-reactive protein; and complete records of nursing interventions such as dressing changes, debridement, dressing selection, and drug treatment.

[0052] Step S102: Perform consistency processing and standardization on the collected raw medical data. First, convert the time information of data from different sources into a standard time format. Then, for individual missing values ​​in the numerical laboratory indicator data stream, analyze the overall trend and periodic characteristics of its time series, and generate reasonable imputation values ​​based on this. For missing values ​​in categorical variables, use the most frequent value under that variable to fill in the missing values. At the same time, parse the wound description text and extract key medical features.

[0053] Step S103: Input the cleaned and standardized complete data into a pre-trained deep learning prediction model. The model adopts a multi-layer neural network structure and can process mixed types of clinical data. After receiving the input, the model performs internal calculations and outputs several candidate care plans for the current patient condition, as well as the predicted wound healing probability and the expected healing time range for each plan.

[0054] Step S104: Analyze the prediction results output by the model, combine them with the medical resource cost parameters set in the system, conduct a comprehensive analysis of each candidate nursing plan, and generate a comparative evaluation report that includes expected effects, time costs, and resource consumption.

[0055] Step S105: Store the generated assessment report in a dedicated medical decision support system database and provide medical staff with a graphical user interface for querying, browsing and comparing assessment details of different options;

[0056] Step S106: Continuously track the actual nursing care plan adopted by the patient and its subsequent clinical course, record the actual wound healing status and time, and link and store these results data in the database to form a closed-loop feedback.

[0057] This embodiment is implemented across multiple wound care centers within a large medical group. The group has established a unified clinical data platform to centrally store and manage patient information from each center. This assessment system, as an advanced application module on this platform, aims to provide clinicians with data-driven decision support. The system adopts a microservice architecture, primarily including data acquisition services, data preprocessing services, model inference services, assessment report generation services, and data storage services. These services communicate with each other through standardized application programming interfaces (APIs) to ensure system scalability and stability.

[0058] The complete workflow of this method is meticulously executed according to the following six core steps:

[0059] Data Acquisition (corresponding to S101): The system is deeply integrated with the medical group's electronic medical record system. Every morning, the system automatically triggers a data acquisition task, extracting the latest data from all registered patients with infected tophi ulcers through a pre-configured data interface. The scope of the collected data is strictly defined, including: basic information from the patient's master index (date of birth, gender, unique medical record number); daily wound assessment data recorded by wound therapists using standardized forms (including length, width, and depth dimensions obtained using professional measuring tools, exudate properties described according to standard classifications (serous, bloody, purulent, etc.), descriptions of surrounding skin redness and swelling, and maceration extent); results of all relevant laboratory tests in the laboratory information system (including serum uric acid concentration, white blood cell count and differential in complete blood count, C-reactive protein, erythrocyte sedimentation rate, and other inflammatory indicators, as well as renal function-related indicators); and every nursing operation recorded in the nursing management system (operation type such as debridement, dressing change, operation time, executor, specific dressing model and brand used, and name and dosage of topical medication). All data is accompanied by a precise timestamp upon extraction.

[0060] Data preprocessing (corresponding to S102): The collected raw data is fed into a multi-stage data preprocessing pipeline. First, time standardization is performed, converting all time information to Coordinated Universal Time (UTC) format and calibrating the time zone. Then, for potential missing values ​​in numerical data streams (such as laboratory indicators), the system analyzes the historical variation patterns and collection frequency of these indicators for the patient, employing an intelligent imputation strategy based on time-series contextual features to generate substitute values. For missing categorical variables (such as exudate type), the most common value under that variable is used for imputation. For wound description text, the preprocessing module initiates a natural language understanding subroutine for deep parsing.

[0061] Model Prediction (corresponding to S103): Rigorously cleaned and standardized data is assembled into a complete data package, which is then invoked in real-time via the application programming interface (API) of the model inference service. This service loads a deep learning prediction model trained on large-scale historical data. Upon receiving input, the model performs a series of complex nonlinear transformations and feature combination calculations within its multi-layered network structure, ultimately generating a structured output: listing several standard care protocols conforming to clinical guidelines, calculating a probability value representing the likelihood of success for each protocol, and an estimated timeframe for wound healing.

[0062] Report Generation (corresponding to S104): After receiving the model's prediction results, the decision engine immediately accesses the medical resource cost knowledge base integrated within the system. This knowledge base contains verified cost parameters. For each candidate treatment plan, the engine calculates a total resource consumption estimate based on its predicted treatment duration and required operations and materials. Subsequently, the engine comprehensively considers the expected effect (healing probability), time cost (healing time), and economic cost (resource consumption) of each plan to generate a detailed comparative evaluation report. The report presents the advantages and disadvantages of each plan in a clear structure.

[0063] Storage and Display (corresponding to S105): The generated assessment reports are persistently stored in a dedicated medical decision support database, which has an optimized storage structure for such reports. Simultaneously, the system's front-end application module provides healthcare professionals with a responsive web-based visual interface. Authorized healthcare professionals can use this interface to retrieve all assessment reports generated for a patient by entering the patient's medical record number or name. The interface displays comparisons of prediction results from different approaches in intuitive formats such as tables and charts.

[0064] Feedback Updates (corresponding to S106): The system implements strict closed-loop management. Regardless of whether the patient ultimately adopts the system-recommended treatment plan or another plan chosen by the doctor, nurses must continuously enter wound assessment records after each nursing care session into the system. When the wound finally heals or the patient is discharged, the outcome event must be marked. The system's backend service automatically and precisely correlates these real, timestamped outcome data with the patient status data used in the initial prediction and the assessment report generated at that time. Then, it stores this complete feedback record in the database, providing high-quality data fuel for the continuous iterative optimization of the model.

[0065] Furthermore, step S102 also includes:

[0066] During structured processing, natural language processing is performed on the ulcer wound feature records to extract key medical entities and convert them into standardized medical terminology codes.

[0067] In this embodiment, step S102 involves parsing the wound description text. Specifically, an entity recognition module based on medical dictionary and natural language processing technology is used to automatically identify key terms such as "wound area," "granulation tissue," and "necrotic tissue" from the free text and map them onto the international standard medical terminology coding system to achieve the structuring and standardization of text information.

[0068] In this embodiment, during the data preprocessing stage, the parsing of wound feature text is performed by a dedicated clinical natural language processing engine. This engine integrates a medical dictionary, medical terminology standards (such as SNOMEDCT), and grammatical analysis rules. It can automatically scan text descriptions entered by nurses, such as: "Right lateral ankle ulcer, 4cm x 3cm, deep subcutaneous, 50% yellow necrotic tissue, moderate amount of purulent exudate, surrounding skin redness and swelling++", and identify key clinical entities, such as "anatomical location: right lateral ankle", "size: 4x3cm", "depth: subcutaneous", "necrotic tissue: necrotic tissue -50%", "exudate: purulent - moderate amount", and "surrounding skin: redness and swelling degree 2+". Subsequently, the engine accurately maps these extracted entities to standard medical terminology codes, thereby achieving accurate conversion from unstructured text to structured data, laying a solid foundation for subsequent model analysis.

[0069] Furthermore, step S103 also includes:

[0070] Reference Figure 2 When constructing the pre-trained nursing effect prediction model:

[0071] Step S301: Collect historical medical record data of patients with tophi ulcers from the medical data warehouses of multiple medical institutions, including basic patient characteristics, ulcer severity scores, inflammatory marker test results, microbial culture results, detailed records of implemented nursing plans, and objective assessment results of wound healing;

[0072] Step S302: Clean and standardize the collected historical gouty tophi ulcer patient medical records, unify ulcer staging standards, laboratory indicator units and nursing operation terminology from different sources, and establish a cross-institutional standard data model;

[0073] Step S303: Construct a multi-task deep learning network architecture, including a feature extraction shared layer and multiple task-specific output layers, to simultaneously predict wound healing probability, healing time, and risk of complications;

[0074] Step S304: Divide the standardized historical gouty tophi ulcer patient medical records into training set, validation set and test set in chronological order, optimize network parameters using a batch training strategy, and use a weighted loss function to balance the importance of different tasks during training.

[0075] Step S305: Monitor model performance on the validation set, use an early stopping strategy to prevent overfitting, and finally evaluate the model's prediction accuracy and generalization ability on the test set.

[0076] Step S306: Deploy the performance-compliant model to the production environment and encapsulate it into an application programming interface (API) that can provide real-time prediction services.

[0077] In this embodiment, the method for constructing the pre-trained deep learning prediction model in step S103 includes the following sub-steps:

[0078] Step S301: Extract desensitized medical record data of all patients with infectious tophi ulcers from the clinical data warehouses of multiple collaborating medical institutions over a historical period. The data items must cover the entire process from admission to wound healing or outcome, including detailed daily assessment records, all implemented treatment and nursing measures, and the final prognosis.

[0079] Step S302: The collected historical data is rigorously cleaned and integrated to unify any differences that may exist between institutions. For example, different ulcer grading systems are converted into a unified staging standard, the units of various laboratory indicators are unified into international standard units, and all nursing operation names are standardized into codes in the standard operation terminology library.

[0080] Step S303: Design a deep neural network model for multi-task learning. This model has a common feature encoder to process all input features, and its downstream is connected to multiple independent output modules, which are respectively responsible for the binary classification task of predicting whether the wound has healed, the regression task of predicting the number of days required for healing, and the classification task of predicting whether a specific complication will occur.

[0081] Step S304: Divide the prepared large-scale dataset into a training subset for training model parameters, a validation subset for adjusting hyperparameters during training, and a test subset for final evaluation of model performance, according to the patients' admission time. Use the training subset data to continuously adjust the weight parameters inside the model through an iterative optimization algorithm.

[0082] Step S305: During model training, continuously use the validation subset to monitor the model's performance on unseen data. When performance no longer improves, terminate training early to prevent overfitting. Finally, use a completely independent test subset to conduct a comprehensive and objective performance evaluation of the final model to assess its prediction accuracy.

[0083] Step S306: Package the evaluated model and all its parameter configurations, deploy them to the model inference engine of the production server, and encapsulate them into a standardized network service interface that can be called by other systems.

[0084] The construction and deployment of the core prediction model is a systematic project, and its entire process includes the following highly detailed steps:

[0085] Historical Data Collection (S301): During the system development phase, after ethical review and approval, complete desensitized data from the data warehouses of three large teaching hospitals within the medical group were extracted from all patients with clearly diagnosed and fully documented infectious tophi ulcers over the past eight years. The data scope covers the entire treatment process from the patient's first visit until wound healing, amputation, or transfer, including detailed daily observation records, all implemented treatment interventions and nursing care measures, and the final clear outcome.

[0086] Data Cleaning and Integration (S302): This involves rigorously cleaning, transforming, and standardizing the massive amounts of historical data collected. This includes: converting various ulcer grading systems used by different hospitals and at different times, such as the Wagner and Texas grading systems, into a unified standard staging framework; standardizing all units of laboratory test indicators to the International System of Units (SI); and mapping all synonymous or near-synonymous nursing procedure names recorded by medical staff, such as "mechanical debridement," "curling debridement," and "sharp instrument debridement," to unique codes in a standard nursing procedure terminology database to ensure data consistency.

[0087] Model Architecture Design (S303): An end-to-end multi-task deep learning model was designed. This model adopts a shared underlying design, where all input features first undergo high-order feature extraction through a common feature encoding network. On top of this, three independent output head branches are constructed: the first branch handles the binary classification task, outputting the probability of wound healing; the second branch handles the regression task, outputting the predicted healing time; and the third branch handles the multi-classification task, outputting the probability of various common complications.

[0088] Model Training and Tuning (S304): The standardized complete dataset is split according to the patients' admission time. The last 20% of the data is used as the test set, the middle 20% as the validation set, and the first 60% as the training set. Using the training set data, the large number of connection weight parameters within the model are iteratively optimized through the error backpropagation algorithm to minimize the overall loss between the model's predicted values ​​and the true labels.

[0089] Model Validation and Evaluation (S305): During training, the model's performance is evaluated on the validation set after each training epoch to monitor for overfitting. Training is terminated early when the validation set performance no longer improves. Finally, a comprehensive and impartial performance evaluation of the final model is performed using a completely independent test set to assess its accuracy and generalization ability across various prediction tasks.

[0090] Model Deployment and Service-Oriented Architecture (S306): The evaluated model and its complete parameter configuration are packaged, converted into a high-performance inference engine format, and deployed to a server cluster in the production environment. Finally, the engine is encapsulated into a standard, remotely invoked RESTful web service interface, complete with full interface documentation, for integration and use by clinical information systems.

[0091] Furthermore, referring to Figure 3 Step S106 is followed by model optimization and updating:

[0092] Step S401: Periodically extract newly accumulated nursing plan execution data and wound healing outcome data from the medical decision support database;

[0093] Step S402: Perform quality control and standardization on the newly extracted data to ensure that it is consistent with the training data format, and label the data with timestamps;

[0094] Step S403: Employ an online learning algorithm to incrementally train the deployed nursing effect prediction model using new data and dynamically adjust the model parameters;

[0095] Step S404: Detect the prediction effect of the model after incremental training through the model performance monitoring system, and deploy a new version of the model when the performance improvement exceeds the set threshold;

[0096] Step S405: Record the metadata for each model update, including update time, data volume, performance change metrics, and store it in the model version management database.

[0097] In this embodiment, step S401: The system sets a fixed model review and update cycle, and automatically retrieves all newly added patient care process and outcome data from the decision support database since the last update after the cycle is reached.

[0098] Step S402: Perform data cleaning, format conversion, and standardization processes on these new data that are completely consistent with the initial training data to ensure data quality and format consistency, and mark each data point with its acquisition timestamp;

[0099] Step S403: Using online learning technology, new data is input into the currently deployed prediction model in small incremental batches. The model parameters are fine-tuned without complete retraining, so that the model can adapt to the new data pattern.

[0100] Step S404: After the model completes incremental learning, use a newly generated validation dataset that the model has never encountered before to evaluate its performance changes. Only after confirming that the model's prediction accuracy has significantly and stably improved will it be updated to the new version for the production environment.

[0101] Step S405: Fully record all key information in this model update process, including the date and time of the update, the quantity and characteristics of the new data used, a detailed performance evaluation report, etc., and store this metadata in the model management log library for version tracking and auditing.

[0102] The system has established a systematic model iteration and update mechanism to ensure that the model can adapt to advancements in medical practice. This mechanism runs automatically according to the following cycle:

[0103] New Data Extraction (S401): The system automatically triggers a model update task every three months. After the task is started, the data pipeline retrieves data from the medical decision support database for all new patient cases that have completed their treatment cycle (i.e., have a clear outcome) since the last update.

[0104] Incremental data preparation (S402): For these new data, perform the same data cleaning, standardization, and feature engineering processes as in the initial model training phase to ensure absolute consistency in data quality and format. Simultaneously, timestamp each data entry to indicate the time it entered the system.

[0105] Online Incremental Learning (S403): This method uses online learning technology to input preprocessed new data into the predictive model deployed in the current production environment in small batches and continuously, fine-tuning the model parameters. The goal of this process is to enable the model to absorb and learn new patterns and knowledge contained in the new data without forgetting existing knowledge.

[0106] Performance Validation and Deployment (S404): After incremental learning is complete, the system automatically uses a recently generated subset of data that the model did not encounter during the incremental learning process as a validation set to evaluate the performance of the updated model. Only when the new model is confirmed to have a significant and stable improvement over the old version across all key metrics will it be approved for deployment to the production environment, replacing the old model.

[0107] Version control and auditing (S405): All operations and results throughout the entire model update lifecycle, including update initiation time, amount of new data processed, performance comparison report of old and new models, operator and other metadata, will be automatically recorded by the system and stored in a dedicated model version management database to realize full lifecycle management and audit tracking of model assets.

[0108] Furthermore, the medical resource consumption parameters mentioned in step S104 include nursing hours, dressing consumption, drug costs, and equipment usage costs.

[0109] In this embodiment, the medical resource cost parameter in step S104 is a pre-configured digital form that defines in detail the standard working hours for each nursing operation, the unit price of various dressings and medicines, and the pricing cost of related medical equipment.

[0110] The decision engine integrates a dynamically updated knowledge base of medical resource cost parameters. This knowledge base is actually a structured database table, jointly maintained by hospital management departments such as the finance department, nursing department, and consumables inventory. Each record details the standard time (in minutes) for a specific nursing procedure, the latest purchase price (in yuan) for a specific medical material, such as dressings or medications, and the billing cost for each use of related medical equipment. When evaluating a treatment plan, the decision engine automatically queries this knowledge base based on the plan's content to accurately calculate the total resource cost required for that plan within the expected treatment period.

[0111] Furthermore, the multi-task deep learning network architecture described in step S303 also includes an attention mechanism module to automatically identify clinical feature variables that are important to the prediction results.

[0112] In this embodiment, the multi-task learning deep neural network model in step S303 also embeds an attention mechanism computation layer. This layer can automatically learn and quantify the importance weight of different input clinical features to the final prediction result, thereby helping the model focus on key information.

[0113] In the internal architecture of the multi-task learning model, an attention mechanism layer is specifically added after the shared feature layer. This layer automatically learns and assigns a weight score to each input feature, representing the importance of that feature in making an accurate prediction. For example, it might assign higher weights to "baseline C-reactive protein level" and "percentage of necrotic tissue in the wound," while assigning lower weights to "patient gender." This not only improves model performance to some extent, but more importantly, it provides insight into the model's decision-making process, enhancing the model's interpretability.

[0114] Furthermore, the visual query interface described in step S105 provides multi-dimensional data analysis functions, including filtering and comparing the effectiveness evaluation results of different nursing plans based on conditions such as ulcer severity, patient age group, and comorbidity type.

[0115] In this embodiment, the graphical user interface in step S105 provides interactive data filtering and comparison functions. Medical staff can select conditions such as the patient's ulcer stage, age group, and whether or not they have diabetes. The system will dynamically filter and display the evaluation results of different nursing plans that meet the conditions side by side, which is convenient for intuitive comparison.

[0116] The system's web-based front-end interface for healthcare professionals includes powerful interactive data insight capabilities. The interface provides a series of dropdown menus and checkboxes as filters, allowing healthcare professionals to dynamically filter data based on multiple dimensions (such as ulcer severity stage, patient age group, and whether they have specific comorbidities like diabetes). Once the filter criteria are set, the system instantly retrieves all matching patient cases from the historical database and displays them side-by-side in a comparison dashboard. This dashboard shows the actual effectiveness statistics of different nursing interventions used in these cases (such as average healing time, healing rate, and complication rate) as well as the model's current predictive assessment values, greatly facilitating healthcare professionals' comparison and selection of interventions based on real-world evidence.

[0117] Furthermore, step S301 also includes a process for de-identifying sensitive patient information, where all personally identifiable information is replaced with anonymous identifiers before model training.

[0118] In this embodiment, the graphical user interface in step S105 provides interactive data filtering and comparison functions. Medical staff can select conditions such as the patient's ulcer stage, age group, and whether or not they have diabetes. The system will dynamically filter and display the evaluation results of different nursing plans that meet the conditions side by side, which is convenient for intuitive comparison.

[0119] The first crucial step after collecting raw data from partner hospitals and before it's used for model training is rigorous data anonymization and desensitization. This process is automated, scanning all data fields and permanently deleting or replacing all direct identifiers (such as names, ID numbers, phone numbers, and detailed addresses) with dummy values. Potential indirect identifiers (such as very precise ages combined with rare diagnoses or specific occupations) are generalized or obfuscated (e.g., converting precise ages into age groups of 5 years). Ultimately, all datasets used for model training and development are completely anonymized, untraceable to any specific individual, and strictly protect patient privacy.

[0120] Furthermore, the nursing effect prediction model described in step S103 also outputs a complication risk score, and in step S104, the safety of the nursing plan is assessed based on the complication risk score.

[0121] The stored procedure in step S105 uses a distributed database system and ensures patient privacy and security through data encryption and access control mechanisms.

[0122] In this embodiment, after collecting data from the medical institution, step S301 immediately performs a strict data desensitization process, which ensures that all data used for model training is anonymized by finding, removing or replacing all information fields that can directly or indirectly identify the patient's personal identity.

[0123] The prediction model in step S103 will also output a potential risk score for complications such as worsening wound infection or damage to surrounding tissues. In step S104, this risk score will be used as an important basis for assessing the safety of the care plan when generating the assessment report.

[0124] The data storage process in step S105 uses a distributed database cluster with high availability and fault tolerance. All assessment report data and patient outcome data written to the database are encrypted and have strict role-based access control policies to ensure the security and compliance of data storage and access.

[0125] In the model's multi-task architecture, the third output branch is dedicated to safety assessment. This branch outputs one or more values ​​between 0 and 1, representing the predicted risk probability of a patient experiencing various specific complications (such as worsening wound infection, osteomyelitis, severe maceration or injury to surrounding skin) during care. The decision engine uses these risk scores as a core safety indicator when generating the final assessment report. For options with predicted risk scores exceeding a preset safety threshold, the report will display prominent warning labels and prompts, reminding healthcare professionals to carefully evaluate the risk-benefit ratio of the option and consider enhanced monitoring or selecting safer alternatives.

[0126] All data storage services in the system are built on a highly available distributed database cluster. This cluster employs a multi-node deployment with automatic failover and data redundancy replication mechanisms, ensuring service continuity and data security. All data (including patient information, assessment reports, model versions, etc.) is encrypted using strong encryption algorithms before being written to the database. Simultaneously, the system implements a role-based, fine-grained access control policy, strictly defining the data access permissions for different users (such as attending physicians, nurses, researchers, and administrators). All data access operations are recorded in an immutable audit log, ensuring the entire system complies with the highest standards and regulatory requirements for medical data security and privacy protection.

[0127] According to a second embodiment of the present invention, the present invention claims protection for an assessment system for the care of infected gouty tophi ulcer wounds, comprising:

[0128] A memory having stored one or more programs that, when executed by one or more processors, cause the one or more processors to implement the aforementioned method for evaluating the care of infected gouty tophi ulcer wounds.

[0129] This system underwent a full-year prospective field trial at two wound care centers within the medical group. During the trial, all newly admitted patients with infected tophi that met the study criteria were included in the trial cohort. Their data were entered into the system in real time to generate assessment reports for the treatment team's reference.

[0130] Systematic test results:

[0131] Predictive accuracy: At the end of the testing period, all predictions (probability of healing, healing time) made by the system for all tested patients throughout the year were compared and statistically analyzed against the actual clinical outcomes observed. Results showed a high degree of consistency between the model's predicted healing probability and the actual healing outcomes. The difference between the model's predicted healing time range and the median actual healing time remained low, demonstrating the effectiveness of the predictions.

[0132] Clinical Applicability: Analysis of physician workstation logs and user interviews revealed that clinicians are increasingly relying on the quantitative assessment reports generated by the system when dealing with complex and refractory wounds. The multi-dimensional comparisons provided in the reports help treatment teams reach a consensus on decisions more quickly, reducing uncertainty.

[0133] Resource management optimization: Comparing operational data from one year before and after the testing period revealed that, in the testing center, the average cost per unit wound healing for patients with infected gouty tophi showed a manageable trend. This indicates that the system's cost assessment function facilitated the adoption of more cost-effective solutions.

[0134] System robustness: During the year-long testing period, the system demonstrated high stability and reliability in the face of various challenges in the clinical environment (such as network fluctuations, data interface changes, and concurrent access), and the availability of core services reached an extremely high level.

[0135] Model evolution capability: During the testing period, the model completed several quarterly incremental updates as planned. After each update, the model was validated using the latest data, and a continuous improvement in predictive performance was observed, demonstrating the effectiveness of the closed-loop feedback and online learning mechanisms.

[0136] Security and Compliance: The entire testing process successfully passed multiple audits by the medical group's information security department. All patient data was properly protected at every stage of processing and storage, and no information security or patient privacy breaches occurred, fully complying with the Personal Information Protection Law and medical industry data security management regulations.

[0137] This detailed embodiment, through the extremely detailed description above, fully demonstrates the implementation details and effects of each technical step of the invention method, from data acquisition to model iteration, fully proving its inventiveness, advancement, and practicality. It completely achieves the intended purpose of the invention and provides a reusable example for the industry.

[0138] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0139] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0140] The specific embodiments of the invention have been described in detail above, but they are only examples, and this application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this application. Therefore, all equivalent changes, modifications, and improvements made without departing from the spirit and principles of this application should be covered within the scope of this application.

Claims

1. A method for evaluating the nursing care of infected gouty tophi ulcer wounds, characterized in that, include: Step S101: Collect standardized electronic medical record data of patients with gouty tophi ulcers through the medical information system. The standardized electronic medical record data includes demographic information, ulcer wound characteristic records, laboratory test index sequences, and previous nursing operation records. Step S102: The collected standardized electronic medical record data is structured, the timestamp format is unified, missing numerical laboratory indicators are filled using time series-based interpolation, and categorical variables are filled using mode imputation. Step S103: Input the preprocessed standardized electronic medical record data into the pre-trained nursing effect prediction model, and use a deep neural network architecture to output the expected wound healing probability and expected healing time corresponding to different nursing plans. Step S104: Analyze the output of the prediction model and generate a nursing plan evaluation report based on the expected wound healing probability, expected healing time, and medical resource consumption parameters; Step S105: Store the nursing plan evaluation report in the medical decision support database and provide a visual query interface for clinical medical staff; Step S106: Collect the actual nursing plan implemented and its corresponding wound healing results data, and update it to the medical decision support database; Step S103 also includes: Historical data collection: During the system development phase, a full set of desensitized data on patients with infectious tophi ulcers with clear diagnoses and complete medical records was extracted from the data warehouse of the teaching hospitals within the medical group. The data range covers the complete treatment process from the patient's first visit to wound healing, amputation, or transfer to another hospital, including detailed observation records, all implemented treatment interventions and nursing measures, and the final clear outcome. Data cleaning and integration: The massive historical data collected is cleaned, transformed and standardized. Various ulcer grading systems used by different hospitals and at different times are uniformly converted into a standard staging framework. The units of all laboratory test indicators are unified into the International System of Units (SI). All synonyms or near-synonyms of nursing operation names recorded by medical staff are standardized and mapped to unique codes in the standard nursing operation terminology database to ensure data consistency. Model Architecture Design: An end-to-end multi-task deep learning model is designed, adopting a shared underlying design. All input features are first extracted through a common feature encoding network to form three independent output head branches: The first branch is responsible for the binary classification task, outputting the probability of wound healing; The second branch is responsible for the regression task and outputs the predicted healing days; The third branch is responsible for multi-classification tasks, outputting the probability of various common complications occurring; Model training and tuning: The standardized complete dataset is divided into test set, validation set and training set according to the order of patient admission time. Using the training set data, the internal connection weight parameters of the model are iteratively optimized through the error backpropagation algorithm to minimize the overall loss between the model prediction value and the true label. Model Validation and Evaluation: During training, the model's performance is evaluated on the validation set after each training epoch to monitor for overfitting. When the validation set performance no longer improves, training is terminated using an early stopping strategy, and the final model's performance is evaluated using a completely independent test set to assess its accuracy and generalization ability on various prediction tasks. Model Deployment and Service-Oriented Architecture: The evaluated model and all parameter configurations are packaged, converted into an inference engine format, and deployed to a server cluster in the production environment. The engine is then encapsulated into a standard, remotely callable RESTful Web service interface, complete with interface documentation, for integration and use by clinical information systems.

2. The method for evaluating the nursing care plan for infectious gouty tophi ulcers according to claim 1, characterized in that, Step S102 also includes: During structured processing, natural language processing is performed on the ulcer wound feature records to extract key medical entities and convert them into standardized medical terminology codes.

3. The method for evaluating the nursing care plan for infectious gouty tophi ulcers according to claim 1, characterized in that, Step S106 is followed by model optimization and updates: Step S401: Periodically extract newly accumulated nursing plan execution data and wound healing outcome data from the medical decision support database; Step S402: Perform quality control and standardization on the newly extracted data to ensure that it is consistent with the training data format, and label the data with timestamps; Step S403: Employ an online learning algorithm to incrementally train the deployed nursing effect prediction model using new data and dynamically adjust the model parameters; Step S404: Detect the prediction effect of the model after incremental training through the model performance monitoring system, and deploy a new version of the model when the performance improvement exceeds the set threshold; Step S405: Record the metadata for each model update, including update time, data volume, performance change metrics, and store it in the model version management database.

4. The method for evaluating the nursing care plan for infectious gouty tophi ulcers according to claim 1, characterized in that, The medical resource consumption parameters mentioned in step S104 include nursing hours, dressing consumption, drug costs, and equipment usage costs.

5. The method for evaluating the nursing care plan for infectious gouty tophi ulcers according to claim 3, characterized in that, The multi-task deep learning model also includes an attention mechanism module that automatically identifies clinical feature variables that are important to the prediction results.

6. The method for evaluating the nursing care plan for infectious gouty tophi ulcers according to claim 1, characterized in that, The visual query interface described in step S105 provides multi-dimensional data analysis functions, including filtering and comparing the effectiveness evaluation results of different nursing plans based on conditions such as ulcer severity, patient age group, and comorbidity type.

7. The method for evaluating the nursing care plan for infectious gouty tophi ulcers according to claim 3, characterized in that, The historical data collection also includes a process for de-identifying sensitive patient information, where all personally identifiable information is replaced with anonymous identifiers before model training.

8. The method for evaluating the nursing care plan for infectious gouty tophi ulcers according to claim 1, characterized in that, The nursing effect prediction model described in step S103 also outputs a complication risk score, and in step S104, the safety of the nursing plan is assessed based on the complication risk score. The stored procedure in step S105 uses a distributed database system and ensures patient privacy and security through data encryption and access control mechanisms.

9. A system for evaluating the care of infected gouty tophi ulcer wounds, characterized in that, include: A memory having stored one or more programs that, when executed by one or more processors, cause the one or more processors to implement an evaluation method for the care of infected gouty tophi ulcer wounds according to any one of claims 1 to 8.

Images