Early warning and predicting method and system for blood gas index trend imbalance

Abstract

The invention provides a blood gas index trend imbalance early warning prediction method and system. The method comprises the following steps: collecting blood gas index data of a patient; comprehensively scoring the blood gas index data by using a fuzzy logic algorithm, and evaluating the acid-base equilibrium state and oxygenation state of the patient to obtain a comprehensive scoring result; according to the comprehensive scoring result, obtaining a classification result; determining that the classification result comprises a warning or critical patient, and obtaining a risk result; analyzing the change trend and periodicity of the blood gas index data to determine the initial health trend of the patient; unsupervised feature learning and dimensionality reduction are carried out on patients with similar health trends, and key features of the health trends are extracted; and comprehensively evaluating the target health trend of the patient. According to the invention, the health state of the patient can be accurately evaluated in real time.

Description

Technical Field

[0001] This application relates to the field of medical information technology, and in particular to a method and system for early warning and prediction of blood gas index trend imbalance. Background Technology

[0002] With the development of medical technology, real-time monitoring and dynamic assessment of patients' health status in medical environments such as intensive care units (ICUs) have become increasingly important. In particular, for critically ill patients with respiratory diseases, cardiovascular diseases, etc., blood gas parameters (such as pH, partial pressure of oxygen, partial pressure of carbon dioxide, and blood oxygen saturation) are important bases for assessing their physiological status and changes in their condition.

[0003] Currently, clinical assessment of a patient's health status primarily relies on the experience and intuition of medical staff, combined with single-indicator data provided by blood gas analyzers. In addition, some medical institutions are beginning to experiment with simple statistical methods or machine learning models to assist in diagnosis and prediction. For example, linear regression models are used to predict a patient's oxygenation status, or decision tree models are used to classify a patient's health condition.

[0004] Existing assessment methods typically consider only individual blood gas indicators, failing to comprehensively reflect a patient's overall health status and easily overlooking important physiological information. Furthermore, they rely on the experience and intuition of healthcare professionals, making them susceptible to individual differences and leading to unstable and unreliable assessment results. Most existing methods only provide static assessment results, failing to capture the dynamic trends of changes in a patient's health status and making it difficult to promptly detect signs of deterioration or improvement. Based on simple statistical models, they lack in-depth analysis of patients' long-term health trends, making it difficult to accurately predict patients' short-term health risks. Summary of the Invention

[0005] This application provides a method and system for early warning and prediction of blood gas index trend imbalance, which solves the problem that it is difficult to accurately predict the short-term health risks of patients in the prior art.

[0006] In a first aspect, embodiments of this application provide a method for early warning and prediction of blood gas index trend imbalance, including:

[0007] Collect patients' blood gas parameters, including pH value, partial pressure of oxygen, partial pressure of carbon dioxide, and blood oxygen saturation;

[0008] The blood gas index data are comprehensively scored using a fuzzy logic algorithm to obtain a comprehensive score result. The fuzzy logic algorithm defines fuzzy sets and membership functions, and combines uncertainty measures based on evidence theory to transform qualitative medical judgments into quantitative scores, assessing the patient's acid-base balance and oxygenation status to obtain a comprehensive score result.

[0009] Based on the comprehensive score results, the patient's health trend is classified to obtain a classification result, which includes at least a good, warning, or critical condition.

[0010] The classification results are determined to include patients with warnings or critical conditions, and the short-term health risks of the patients are assessed using survival analysis methods to obtain risk results. The survival analysis methods predict the patients' survival time and risk factors using Cox proportional hazards models and Aalen multiplicative models and use these predictions as risk results.

[0011] The long short-term memory network model is used to capture the long-term dependencies in the patient's historical blood gas index data, and combined with the comprehensive score results, the changing trends and periodicity of the blood gas index data are analyzed to determine the patient's preliminary health trend.

[0012] Based on the patients' preliminary health trends, unsupervised feature learning and dimensionality reduction are performed on patients with similar health trends, and key features of health trends are extracted.

[0013] Based on the classification results, the risk results, and the key features of the health trends, the patient's target health trends are comprehensively evaluated.

[0014] Secondly, embodiments of this application provide a blood gas index trend imbalance early warning and prediction system, including:

[0015] The data acquisition module is used to collect patients' blood gas parameters, including pH value, partial pressure of oxygen, partial pressure of carbon dioxide, and blood oxygen saturation.

[0016] The calculation module is used to perform a comprehensive score on the blood gas index data using a fuzzy logic algorithm to obtain a comprehensive score result. The fuzzy logic algorithm defines fuzzy sets and membership functions, and combines them with an uncertainty measure based on evidence theory to transform qualitative medical judgments into quantitative scores, assess the patient's acid-base balance and oxygenation status, and obtain a comprehensive score result.

[0017] The classification module is used to classify the patient's health trend based on the comprehensive score result, and obtain the classification result, which includes at least good, warning or critical.

[0018] An analysis module is used to identify patients whose classification results include warnings or critical conditions, and to assess the short-term health risks of the patients using survival analysis methods to obtain risk results. The survival analysis methods use a Cox proportional hazards model and an Aalen multiplicative model to predict the patients' survival time and risk factors and use these as risk results.

[0019] The determination module is used to capture long-term dependencies in the patient's historical blood gas index data using a long short-term memory network model, and to analyze the changing trends and periodicity of the blood gas index data in combination with the comprehensive score results, so as to determine the patient's preliminary health trend.

[0020] The extraction module is used to perform unsupervised feature learning and dimensionality reduction on patients with similar health trends based on the patient's preliminary health trends, and to extract the key features of the health trends.

[0021] The assessment module is used to comprehensively evaluate the patient's target health trend based on the classification results, the risk results, and the key features of the health trend.

[0022] Thirdly, embodiments of this application provide a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are to be invoked and executed by the processing component to implement a blood gas index trend imbalance early warning and prediction method as described in the first aspect above.

[0023] Fourthly, embodiments of this application provide a computer storage medium storing a computer program, which, when executed by a computer, implements a method for early warning and prediction of blood gas index trend imbalance as described in the first aspect.

[0024] In this embodiment, blood gas parameters (such as pH, oxygen partial pressure, carbon dioxide partial pressure, and blood oxygen saturation) are collected from patients. A fuzzy logic algorithm is used to perform a comprehensive scoring, transforming qualitative medical judgments into quantitative scores to assess patients' acid-base balance and oxygenation status. Based on the comprehensive scoring results, patients' health trends are classified (good, warning, or critical), and survival analysis is used to assess the short-term health risks of patients in the warning or critical stage. Simultaneously, a long short-term memory network model is used to analyze the changing trends and periodicity of blood gas parameters. Unsupervised feature learning and dimensionality reduction techniques are combined to extract key features of health trends. Finally, based on the classification results, risk assessment, and key features, the patient's health trend is comprehensively evaluated, achieving accurate assessment, personalized health management, and risk prediction.

[0025] Furthermore, by employing fuzzy logic algorithms and uncertainty metrics, complex physiological states are transformed into accurate quantitative scores, improving the reliability and accuracy of the scores. Through ensemble learning methods using multiple decision trees, personalized classification of patient health trends is achieved, helping doctors quickly identify patients requiring special attention. Research has found that the application of these technologies is particularly significant in emergency and intensive care settings.

[0026] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 A flowchart illustrating a method for early warning and prediction of blood gas index trend imbalance provided in this application embodiment;

[0029] Figure 2 A schematic diagram of blood gas index data monitored in a patient in a blood gas index trend imbalance early warning and prediction method provided in an embodiment of this application;

[0030] Figure 3 A schematic diagram of the structure of a blood gas index trend imbalance early warning and prediction system provided in this application embodiment;

[0031] Figure 4 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0032] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0033] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0034] 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 some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0035] Figure 1 A flowchart of a method for early warning and prediction of blood gas index trend imbalance is provided in this application embodiment, as shown below. Figure 1 As shown, the method includes:

[0036] 101. Collect the patient's blood gas parameters, including pH value, partial pressure of oxygen, partial pressure of carbon dioxide and blood oxygen saturation;

[0037] In this step, blood gas parameters refer to a series of data obtained from the patient's blood sample that reflect respiratory function and acid-base balance, mainly including pH value, partial pressure of oxygen (PaO2), partial pressure of carbon dioxide (PaCO2), and blood oxygen saturation (SpO2).

[0038] In this embodiment of the application, blood samples from patients are collected using medical equipment, and blood gas analyzers are used to measure the specific values ​​of the above-mentioned indicators. These data will be used for subsequent comprehensive scoring and health trend analysis.

[0039] 102. The blood gas index data are comprehensively scored using a fuzzy logic algorithm to obtain a comprehensive score result. The fuzzy logic algorithm defines fuzzy sets and membership functions, and combines uncertainty measures based on evidence theory to transform qualitative medical judgments into quantitative scores, assessing the patient's acid-base balance and oxygenation status to obtain a comprehensive score result.

[0040] In this step, fuzzy logic algorithms are mathematical tools for handling uncertainty and fuzziness. By defining fuzzy sets and membership functions, qualitative medical judgments are transformed into quantitative scores.

[0041] In this embodiment, fuzzy sets are defined based on the physiological characteristics of acid-base balance and oxygenation status. Membership functions are established, and the membership degree of each blood gas index data in each fuzzy set is calculated. Combining an uncertainty measure based on evidence theory, the uncertainties and conflicts between different indicators are integrated to generate a comprehensive score result for assessing the patient's acid-base balance and oxygenation status.

[0042] Optionally, step 102, "using fuzzy logic algorithm to comprehensively score the blood gas index data and obtain a comprehensive score result," includes: defining relevant fuzzy sets based on the physiological characteristics of acid-base balance and oxygenation status, each fuzzy set representing a specific physiological state; establishing membership functions using the defined fuzzy sets, the membership functions determining the membership degree of the corresponding fuzzy sets based on the actual measured values ​​of the blood gas index data; processing the blood gas index data based on the established membership functions to calculate the membership degree of the fuzzy set corresponding to each blood gas index data; integrating the uncertainties and conflicts between different blood gas index data using the membership degree of the fuzzy set corresponding to each blood gas index data, combined with an uncertainty measure based on evidence theory, to obtain an uncertainty measure result; converting the acquired qualitative medical judgment for the patient into a quantitative score based on the uncertainty measure result, and generating a comprehensive score result based on the quantitative score, the comprehensive score result being used to assess the patient's acid-base balance and oxygenation status.

[0043] Fuzzy sets: Fuzzy sets are a mathematical tool for handling uncertainty and fuzziness, used to describe concepts with unclear boundaries. In this method, multiple fuzzy sets are defined based on the physiological characteristics of acid-base balance and oxygenation states. Each fuzzy set represents a specific physiological state, such as "normal," "mildly abnormal," or "severely abnormal."

[0044] Membership function: The membership function defines the degree to which a specific data point belongs to a certain fuzzy set, and its value is usually between 0 and 1. The membership function determines the membership degree of the data point in each fuzzy set based on the actual measured value of the blood gas index data.

[0045] Uncertainty Measurement: Uncertainty measurement based on evidence theory is used to handle and integrate uncertainties and conflicts between different blood gas index data. Evidence theory provides a comprehensive uncertainty measurement result by combining information from multiple evidence sources.

[0046] Quantitative scoring: converting qualitative medical judgments (such as "mild acidosis" or "severe hypoxia") into specific numerical scores to facilitate further analysis and comparison.

[0047] First, based on the physiological characteristics of acid-base balance and oxygenation, multiple fuzzy sets are defined, each representing a specific physiological state. For example, pH value can be defined as "normal" (7.35-7.45), "mild acidosis" (7.30-7.35), and "severe acidosis" (<7.30). A membership function is established for each fuzzy set, and the membership degree of the data point in each fuzzy set is determined based on the actual measured values ​​of blood gas indicators. For example, when the pH value is 7.32, the membership degree in the "normal" set might be 0.4, and the membership degree in the "mild acidosis" set might be 0.6. Next, the established membership functions are used to process each blood gas indicator data, calculating its membership degree in each fuzzy set. Further, an uncertainty measurement method based on evidence theory is used to integrate the uncertainties and conflicts between different blood gas indicator data, obtaining a comprehensive uncertainty measurement result. Finally, based on the uncertainty measurement result, qualitative medical judgments are transformed into quantitative scores, and a comprehensive score result is generated based on these quantitative scores. The comprehensive score is used to assess the patient's acid-base balance and oxygenation status.

[0048] In this embodiment of the application, it is assumed that a patient has the following blood gas parameters: pH value: 7.32, partial pressure of oxygen (PaO2): 70 mmHg, partial pressure of carbon dioxide (PaCO2): 45 mmHg, and blood oxygen saturation (SpO2): 92%.

[0049] Fuzzy sets are defined as follows: pH value: normal (7.35-7.45), mild acidosis (7.30-7.35), severe acidosis (<7.30). PaO2: normal (>80 mmHg), mild hypoxia (60-80 mmHg), severe hypoxia (<60 mmHg). PaCO2: normal (35-45 mmHg), mild hypercapnia (45-50 mmHg), severe hypercapnia (>50 mmHg). SpO2: normal (>95%), mild hypoxia (90-95%), severe hypoxia (<90%). Membership functions are established: when the pH value is 7.32, the membership degree is: normal (0.4), mild acidosis (0.6). When the PaO2 is 70 mmHg, the membership degree is: normal (0.3), mild hypoxia (0.7). When PaCO2 is 45 mmHg, the membership degree is: normal (0.5), mild hypercapnia (0.5). When SpO2 is 92%, the membership degree is: normal (0.6), mild hypoxia (0.4).

[0050] Membership was calculated as follows: pH: Normal (0.4), Mild acidosis (0.6); PaO2: Normal (0.3), Mild hypoxia (0.7); PaCO2: Normal (0.5), Mild hypercapnia (0.5); SpO2: Normal (0.6), Mild hypoxia (0.4).

[0051] Using evidence theory, the aforementioned membership degrees are integrated to obtain a comprehensive uncertainty measure. For example, considering all indicators, the patient's overall condition might be assessed as "mildly abnormal." Based on the uncertainty measure, the qualitative medical judgment of "mildly abnormal" is converted into a quantitative score, for example, a score range of 0-100, where mildly abnormal might score 30. The final comprehensive score of 30 indicates that the patient's acid-base balance and oxygenation status are mildly abnormal, requiring further observation and treatment.

[0052] See Figure 2 In a patient monitoring data case; the patient's example data is (pH: 7.32, PaO2: 70 mmHg, PaCO2: 45 mmHg, SpO2: 92%). The membership degree and comprehensive score results calculated using fuzzy logic are detailed below. Figure 2 Finally, based on the comprehensive scoring results, the system outputs a score of 30 points. Ultimately, the system classifies the patient's health trend as "warning," indicating that the patient is currently in a mildly abnormal state and requires close observation and consideration of intervention measures.

[0053] 103. Based on the comprehensive score results, classify the patient's health trend to obtain classification results, which include at least good, warning, or critical.

[0054] In this step, health trend classification categorizes the patient's health status into different categories based on the comprehensive score results, typically including good, warning, and critical.

[0055] In this embodiment, a classification standard for health trends is defined, including preset scoring thresholds. Based on the comprehensive scoring results, patients are categorized into the corresponding health trend categories, enabling doctors to quickly identify patients requiring special attention.

[0056] Optionally, step 103, "classifying the patient's health trends based on the comprehensive scoring results to obtain classification results," includes: defining classification criteria for health trends based on the comprehensive scoring results, wherein the classification criteria include preset scoring thresholds used to distinguish different categories of health trends; constructing multiple decision trees, each decision tree being independently trained by randomly sampling and feature selection from a training dataset, each decision tree performing a preliminary classification of the patient's health trends based on the comprehensive scoring results and the classification criteria; and determining the patient's final health trend classification based on the classification results of the multiple decision trees, and using this classification result.

[0057] Comprehensive score result: This refers to a value obtained by a specific algorithm after a series of health indicators or test results. This value can reflect the patient's overall health status at a certain point in time.

[0058] Classification criteria for health trends: In order to classify patients' health status, a set of classification criteria needs to be defined. This set of criteria usually includes several preset scoring thresholds, each threshold corresponding to a category of health trend, such as "health improvement", "stable", "deterioration", etc.

[0059] Decision tree: A supervised learning method that uses a tree structure to make decisions. Each internal node represents a test on an attribute, each branch represents a test output, and each leaf node represents a classification result.

[0060] Training dataset: This is a collection of patient data containing a large number of patients with known health statuses, used to train the model so that it can learn how to predict patients' health trends based on the input data.

[0061] First, based on professional knowledge and historical data, scoring thresholds for different health trend categories are defined. For example, a comprehensive score of 80 or above is considered "improved health"; 50-80 is "stable"; and below 50 is "deteriorating health". Next, using the training dataset, multiple decision trees are constructed through random sampling and feature selection. Each tree is trained on different samples and feature subsets, which helps improve the model's generalization ability and reduce the risk of overfitting. Second, for each patient to be classified, all the constructed decision tree models are used to classify them separately, and each model provides a preliminary classification result. Finally, the classification results of all decision trees are aggregated, and a majority voting mechanism is used to determine the final health trend classification. For example, if 70% of the decision trees consider the patient's health trend to be "improved", then the final classification result is "improved".

[0062] In this embodiment, it is assumed that a system is being developed for the health management of patients with cardiovascular diseases. The system collects physiological parameters such as blood pressure, heart rate, and blood sugar levels, as well as non-physiological information such as lifestyle habits (e.g., diet and exercise), and calculates a comprehensive score for each patient using a specific algorithm.

[0063] The system defines a comprehensive score of 90 or above as "significantly improved health," 70-90 as "improved health," 50-70 as "stable health," 30-50 as "slightly declining health," and below 30 as "deteriorating health." Next, 10,000 records are randomly selected from cardiovascular disease patient data over the past five years as the training dataset. These data are used to construct 100 decision trees, each trained based on a different subset of data and feature combinations. When new patient data is input, these 100 decision trees are evaluated individually. Assuming 80 decision trees determine improved health, 15 determine stable health, and 5 determine declining health, then according to the majority principle, the system's final classification result will be "improved health."

[0064] 104. Identify patients whose classification results include warnings or critical conditions, and use survival analysis methods to assess the short-term health risks of the patients to obtain risk results. The survival analysis methods use Cox proportional hazards models and Aalen multiplicative models to predict the patients' survival time and risk factors and use them as risk results.

[0065] In this step, short-term health risk assessment is performed on patients whose health condition is rated as warning or critical, using survival analysis methods to predict the patient's survival time and risk factors.

[0066] In this embodiment, the Cox proportional hazards model and the Aalen multiplicative model are used to assess the patient's short-term health risks, predict survival time and risk factors, and provide a basis for clinical decision-making based on the patient's comprehensive score and other relevant data.

[0067] This application considers the critical importance of assessing short-term health risks in the medical field, particularly for patients with warning or critical conditions. This not only helps physicians develop timely and effective treatment plans but also aids in the efficient allocation of hospital resources. Survival analysis is a statistical method that studies the length of time an event occurs, such as the time from diagnosis to death or from surgery to recurrence. Through survival analysis, a patient's future health status can be predicted, leading to more rational medical decisions.

[0068] The specific options are as follows:

[0069] Optionally, step 104, "determining that the classification results include patients with warnings or critical conditions, and using survival analysis to assess the short-term health risks of the patients to obtain risk results," includes: applying a Cox proportional hazards model to predict the survival time and risk factors of patients with warnings or critical conditions, wherein the model expression of the Cox proportional hazards model is: ;

[0070] in, Indicates having covariates Individuals in time The risk function; represents the baseline risk function, and represents the risk function when there are no covariates influencing the risk. Indicates time and covariates A function representing how risk factors change over time and with individual characteristics; Indicates the first One covariate, where a covariate refers to a risk factor; Indicates time The function represents the first and the The impact of the interaction between individual covariates on the risk function; Representing a nonlinear function, it refers to a model based on machine learning. These represent model parameters, used to capture complex nonlinear relationships between covariates; Representing covariates The index; Representing covariates The index; Indicates the total number of covariates;

[0071] The Aalen multiply-add model was used to nonparametrically estimate the survival time of patients with warnings or critical conditions, yielding the instantaneous hazard ratio. The model expression for the Aalen multiply-add model is as follows: ;

[0072] in, Indicates time The instantaneous risk rate; represents the baseline risk function, and represents the risk function when there are no covariates influencing the risk. Indicates time and covariates The function represents the first The impact of each covariate on the risk function; Indicates the first Covariates over time The value of represents the change of the covariate over time; Indicates time The function represents the first and the The impact of the interaction between individual covariates on the risk function; Representing a nonlinear function, it refers to a model based on machine learning. These represent model parameters, used to capture complex nonlinear relationships between covariates; Representing covariates The index; Representing covariates The index; Indicates the total number of covariates;

[0073] Based on the prediction results of the Cox proportional hazards model and the Aalen multiplicative model, the patient's survival time and risk factors are comprehensively assessed to obtain the risk outcome.

[0074] In step 104, it is necessary to assess the short-term health risks of patients under warning or in critical condition. For this purpose, two classic survival analysis methods were selected: the Cox proportional hazards model and the Aalen multiplicative model. Each model has its advantages: the Cox proportional hazards model can handle the relative impact of covariates on risk, while the Aalen multiplicative model better captures the impact of covariates changing over time. Combining the advantages of the Cox proportional hazards model and the Aalen multiplicative model aims to more accurately assess the short-term health risks of patients. The Cox proportional hazards model assumes a constant hazard ratio, while the Aalen multiplicative model allows the hazard ratio to change over time, making the latter more suitable for handling time-varying risk factors. Combining the two allows for the simultaneous consideration of both fixed and time-dependent risk factors, improving predictive accuracy.

[0075] In the Cox proportional hazards model expression, the baseline hazard function This represents the risk function without any covariates, used to capture the effect of time itself on risk. Covariates This represents various risk factors, such as age, gender, and blood pressure, used to assess the impact of these factors on risk. Time dependence coefficient. This indicates how risk factors change over time and with individual characteristics, capturing dynamic impacts. Interactions. This represents the impact of interactions between different covariates on risk, and is used to capture complex interactions. Nonlinear function. Machine learning-based models are used to capture complex nonlinear relationships between covariates, improving the flexibility and accuracy of the models.

[0076] Among them, the baseline risk function The baseline risk function is typically a nonparametric estimate or a simple function, such as: in It is the baseline survival function.

[0077] Time-dependent coefficient These coefficients can be time. and covariates Functions. For example, they can be represented as: in These are model parameters.

[0078] Interaction coefficient These coefficients can be time. A function that represents the interaction between covariates. For example: in These are model parameters.

[0079] nonlinear functions This function can be a complex machine learning model, such as a deep neural network (DNN), and its expression depends on the network's architecture. For example, a simple... It can be represented as: in Indicates model parameters, and These are the weights and biases of the network layers. It is an activation function (such as ReLU or sigmoid).

[0080] In the Aalen multiply-add model expression, the baseline risk function This represents the risk function without any covariates, used to capture the effect of time itself on risk. Covariates This indicates the time period of each risk factor. The value of is used to assess the impact of these factors on risk over time. Time dependence coefficient. Indicates the first The impact of each covariate on the risk function, capturing dynamic effects. Interactions. This represents the impact of interactions between different covariates on risk, and is used to capture complex interactions. Nonlinear function. Machine learning-based models are used to capture complex nonlinear relationships between covariates, improving the flexibility and accuracy of the models.

[0081] Among them, the baseline risk function The baseline risk function is typically a nonparametric estimate or a simple function, such as: in It is the baseline survival function.

[0082] Time-dependent coefficient These coefficients can be time. and covariates Functions. For example, they can be represented as: in These are model parameters.

[0083] Interaction coefficient These coefficients can be time. A function that represents the interaction between covariates. For example: in These are model parameters.

[0084] nonlinear functions This function can be a complex machine learning model, such as a deep neural network (DNN), and its expression depends on the network's architecture. For example, a simple DNN can be represented as: in Indicates model parameters, and These are the weights and biases of the network layers. It is an activation function (such as ReLU or sigmoid).

[0085] The following is a brief explanation of how each parameter is obtained:

[0086] Baseline risk function and It uses the maximum likelihood estimation method to fit the data using survival data without covariates. and It is extracted from the patient's historical data, such as age, gender, blood pressure, and blood sugar. Time-dependent coefficient. and It uses the maximum likelihood estimation method to fit survival data including covariates. (Interaction) and It uses the maximum likelihood estimation method to fit survival data including interaction terms. (Nonlinear function) and It involves fitting parameters using machine learning methods (such as neural networks, random forests, etc.). and Training is performed using optimization algorithms such as gradient descent.

[0087] In this embodiment of the application, it is assumed that the short-term health risks of patients with heart disease are being studied, and a group of patients with health trends of "warning" or "critical" have been obtained through preliminary classification.

[0088] Patient data included age, sex, blood pressure, cholesterol levels, and smoking history. The time from admission to the onset of a cardiac event was recorded for each patient, or the time before the event occurred by the end of the study (censored data).

[0089] Cox proportional hazards model: baseline hazard function Fitting was performed using the maximum likelihood estimation method. Covariates Extracted from patient data. Time-dependent coefficient. Fitting was performed using the maximum likelihood estimation method. Interaction. Fitting is performed using the maximum likelihood estimation method. Nonlinear function. Fitting using a neural network model.

[0090] Aalen multiplicative model: baseline risk function Fitting was performed using the maximum likelihood estimation method. Covariates Extracted from patient data, taking into account changes over time. Time-dependent coefficient. Fitting was performed using the maximum likelihood estimation method. Interaction. Fitting is performed using the maximum likelihood estimation method. Nonlinear function. Fitted using a random forest model.

[0091] Based on the prediction results of the Cox proportional hazards model and the Aalen multiplicative model, the short-term health risk of patients is comprehensively assessed. Short-term risk prediction calculates the probability of each patient experiencing a cardiac event within the next 30 days. Personalized medical intervention recommendations are made based on the risk assessment results. For high-risk patients, it is recommended to strengthen medication treatment, such as increasing the dosage of anticoagulants. Patients are also advised to quit smoking, control their diet, and increase physical activity. Regular cardiac function tests are recommended for high-risk patients to allow for timely adjustments to the treatment plan.

[0092] This comprehensive assessment approach can provide heart disease patients with more precise and personalized risk management advice, helping them reduce short-term health risks and improve their quality of life.

[0093] Furthermore, in the process of comprehensively assessing the patient's short-term health risk based on the prediction results of the Cox proportional hazards model and the Aalen multiplicative model, firstly, from the risk function... The predictions of survival time and risk factors were derived.

[0094] The prediction from risk rate to survival time includes:

[0095] Risk rate It describes a certain moment Given covariates Under certain conditions, the instantaneous rate at which an event occurs. Survival time. This refers to the time until the event occurs (such as death or other health events). Survival function This indicates that the patient has survived beyond the specified time. The probability can be obtained by integrating the risk rate: ;

[0096] here, Indicates time The risk rate, and From time 0 to The cumulative risk. This score gives the patient's survival time. The complement of the probability, that is, the probability that the event occurred before the patient's time. Therefore, This indicates that the patient has survived beyond the specified time. The probability of.

[0097] The prediction of risk rates and risk factors includes:

[0098] Risk factors are variables that affect patient survival time; they are expressed through coefficients in the model (such as...). and These coefficients quantify how each risk factor affects the risk rate.

[0099] Hazard ratios: In the Cox model, the coefficients... index The risk ratio is given, which is the risk ratio for each additional unit of covariate. The risk increases exponentially.

[0100] Quantification of risk factors: Model coefficients provide a quantitative measure of risk factors. For example, if... A positive value indicates An increase in [something] is associated with an increase in risk; if A negative value indicates The increase is associated with a decrease in risk.

[0101] Risk factor identification: Statistical tests (such as the Wald test) can be used to determine which covariates are obvious risk factors.

[0102] Personalized risk prediction uses model coefficients to calculate an individualized risk score for each patient, which helps identify high-risk patients and provide personalized management.

[0103] Through survival function This can predict the distribution of patient survival time. Through model coefficients, key risk factors affecting patient survival time can be identified and quantified. This information aids clinical decision-making, such as resource allocation, treatment planning, and patient monitoring.

[0104] Secondly, according to the risk function in the Aalen multiply-add model Derive predictions of survival time and risk factors. Prediction from hazard ratio to survival time using the hazard function in the Aalen multiplicative model. Described in time The instantaneous risk rate, that is, the rate at which an event occurs at that moment. Survival function. This indicates that the patient has survived beyond the specified time. The probability of survival. The relationship between the survival function and the hazard function can be expressed by the following integral: ;

[0105] here, Indicates time The risk rate, and From time 0 to The cumulative risk. This score gives the patient's survival time. The complement of the probability, that is, the cumulative probability that the event occurred before this time. Therefore, This indicates that the patient has survived beyond the specified time. The probability of.

[0106] From Risk Rate to Risk Factor Prediction: Risk factors are variables that affect patient survival time, and they are predicted through coefficients in the model (such as...). and These coefficients quantify how each risk factor affects the risk rate.

[0107] Quantification of risk factors: Model coefficients provide a quantitative measure of risk factors. For example, if... A positive value indicates An increase in [something] is associated with an increase in risk; if A negative value indicates The increase is associated with a decrease in risk.

[0108] Risk factor identification: Statistical tests (such as likelihood ratio tests) can be used to determine which covariates are significant risk factors.

[0109] Personalized risk prediction: Using model coefficients, an individualized risk score can be calculated for each patient, which helps to identify high-risk patients and provide personalized management.

[0110] Through survival function This can predict the distribution of patient survival time. Through model coefficients, key risk factors affecting patient survival time can be identified and quantified. This information aids clinical decision-making, such as resource allocation, treatment planning, and patient monitoring.

[0111] 105. Using a long short-term memory network model to capture the long-term dependencies in the patient's historical blood gas index data, and combining the comprehensive score results, analyze the changing trends and periodicity of the blood gas index data to determine the patient's preliminary health trend;

[0112] In this step, trend and periodicity analysis refers to capturing long-term dependencies in patients' historical blood gas index data using a long short-term memory network model (LSTM) to analyze the trends and periodic patterns of the data.

[0113] In this embodiment, an LSTM model is used to process the patient's historical blood gas index data, capture the long-term dependencies in the data, and combine the comprehensive scoring results to analyze the changing trends and periodicity of the blood gas index data, thereby determining the patient's preliminary health trend.

[0114] Optionally, step 105, "using a long short-term memory network model to capture the long-term dependencies in the patient's historical blood gas index data, and combining the comprehensive score results to analyze the changing trends and periodicity of the blood gas index data to determine the patient's preliminary health trend," includes: collecting the patient's historical blood gas index data, which includes time-series data of pH value, partial pressure of oxygen, partial pressure of carbon dioxide, and blood oxygen saturation; constructing a long short-term memory network model, using the gating mechanism of the long short-term memory network model to capture the long-term dependencies and short-term patterns in the patient's historical blood gas index data; inputting the historical blood gas index data and the comprehensive score results into the long short-term memory network model to analyze the changing trends and periodicity of the blood gas index data; and determining the patient's preliminary health trend based on the output of the long short-term memory network model, which includes the changing trends of acid-base balance and periodic changes in oxygenation status.

[0115] 106. Based on the patient's preliminary health trend, perform unsupervised feature learning and dimensionality reduction on patients with similar health trends, and extract the key features of the health trend;

[0116] In this step, the key feature of health trends is the important characteristics that can reflect the patient's health status, extracted from a large amount of data.

[0117] In this embodiment, unsupervised feature learning and dimensionality reduction are performed on patients with similar health trends to extract key features of these health trends. These features help to understand the disease progression and prognostic assessment, providing support for clinical research and treatment.

[0118] Optionally, step 106, "based on the patient's preliminary health trend, performing unsupervised feature learning and dimensionality reduction on patients with similar health trends, and extracting key features of the health trends," includes: selecting a group of patients with similar health trend features based on the patient's preliminary health trend; selecting or constructing an unsupervised learning model, the unsupervised learning model including an autoencoder algorithm, for learning feature representations from the data corresponding to the patients with similar health trends; using the unsupervised learning model to perform feature learning and dimensionality reduction on the data corresponding to the patients with similar health trends, in order to extract latent structures and patterns in the data, and map high-dimensional data to a low-dimensional space to obtain dimensionality-reduced data; extracting key features from the dimensionality-reduced data, the key features being used to represent the core changes and patterns of the patient's health trend.

[0119] Preliminary health trend: This is a preliminary classification of the patient's health status based on the comprehensive scoring results and decision tree model from previous steps, such as "improved health", "stable", or "deteriorating".

[0120] Patients with similar health trends: This refers to other patient groups who have the same or very similar classification results as the target patients in terms of initial health trends. This group is selected to better understand the factors and patterns behind specific health trends.

[0121] Unsupervised learning models: These models can automatically discover structures and patterns in data without labeled data. Common unsupervised learning models include clustering algorithms and autoencoders.

[0122] Autoencoder algorithms are neural network architectures designed to learn an effective representation of data (encoding) while reconstructing the original input as faithfully as possible (decoding). Autoencoders are particularly suitable for feature learning and dimensionality reduction tasks.

[0123] Dimensionality reduction: The process of transforming high-dimensional data into low-dimensional data. The purpose is to remove redundant information, retain the essential characteristics of the data, and reduce computational complexity and storage requirements.

[0124] Key features: Features extracted from the dimensionality-reduced data that best describe the core changes and patterns in patient health trends. These features are crucial for subsequent health analysis and prediction.

[0125] First, based on the preliminary health trend classification of the target patients, a group of patients with similar health trends is selected from the database. Next, an unsupervised learning model suitable for feature learning and dimensionality reduction tasks, such as an autoencoder, is selected or constructed. Then, the selected unsupervised learning model is used to learn features from the high-dimensional data of patients with similar health trends, while dimensionality reduction mapping the data to a lower-dimensional space to extract latent structures and patterns. Finally, the key features that best represent the core changes and patterns in the patients' health trends are identified and extracted from the dimensionality-reduced data.

[0126] In this embodiment of the application, it is assumed that the study is on the health trend management of diabetic patients, and a group of patients with a "stable" health trend has been obtained through preliminary classification.

[0127] Patients with similar health trends were selected: 1,000 diabetic patients with a similar "stable" health trend were chosen from the database. These patients had similar blood glucose control, weight changes, and lifestyle habits. An autoencoder was then chosen as the unsupervised learning model. The autoencoder consists of an encoder and a decoder. The encoder compresses high-dimensional patient data into low-dimensional feature vectors, while the decoder attempts to reconstruct the original data from these feature vectors. The autoencoder was then used to train the model on the high-dimensional data of these 1,000 patients (including blood glucose levels, dietary habits, exercise frequency, etc.). The encoder compresses each patient's high-dimensional data into a low-dimensional feature vector, which captures the main patterns of the patient's health trend. Finally, from the low-dimensional feature vectors generated by the autoencoder, further analysis (such as correlation analysis and principal component analysis) identified several key features. For example, it may be found that "postprandial blood glucose fluctuations," "weekly exercise frequency," and "sleep quality" are the most important factors affecting the stability of the health trend in diabetic patients.

[0128] 107. Based on the classification results, the risk results, and the key features of the health trends, comprehensively evaluate the patient's target health trends.

[0129] In this step, the target health trend refers to a comprehensive evaluation of the patient's future health status and development direction based on multiple assessment results.

[0130] In this embodiment, the patient's future health trend is comprehensively evaluated by combining health trend classification results, short-term health risk assessment results, and key characteristics of health trends. This comprehensive evaluation provides doctors with comprehensive data support for developing personalized treatment plans and management strategies.

[0131] Optionally, step 107, "comprehensively evaluating the patient's target health trend based on the classification results, the risk results, and the key features of the health trend," includes: integrating the classification results, the risk results, and the key features of the health trend to form a patient health information database; using a multi-criteria decision analysis method, comprehensively evaluating the patient's long-term and short-term health trends based on the patient health information database to obtain a comprehensive evaluation result; and outputting the patient's target health trend based on the comprehensive evaluation result, including short-term risk prediction, long-term health change trends, and personalized medical intervention suggestions.

[0132] Classification results: refers to the preliminary classification of the patient's health trend based on the comprehensive scoring results and decision tree model, such as "health improvement", "stable" or "deterioration".

[0133] Risk outcome: refers to the probability that a patient may experience health problems in the short term, such as heart attack or diabetic complications, as calculated by a risk assessment model.

[0134] Key features of health trends: Features extracted from dimensionality-reduced data that best describe the core changes and patterns in patient health trends, which are crucial for understanding patient health status.

[0135] Patient health information database: This database integrates key features of classification results, risk outcomes, and health trends to form a comprehensive patient health information database for subsequent comprehensive evaluation and analysis.

[0136] Multi-criteria decision analysis methods: These are analytical methods used to solve multi-objective decision problems. They comprehensively evaluate alternative solutions by considering multiple criteria or factors. Commonly used multi-criteria decision analysis methods include the Analytic Hierarchy Process (AHP) and the Top-Ranking System of Technology (TOPSIS).

[0137] Comprehensive evaluation results: Based on a multi-criteria decision analysis method, combined with various information from the patient's health information database, the results comprehensively evaluate the patient's long-term and short-term health trends.

[0138] Target health trends: The final output of patient health trends, including short-term risk predictions, long-term health change trends, and personalized medical intervention recommendations.

[0139] First, the patient's classification results, risk outcomes, and key characteristics of health trends are integrated into a single database to form a patient health information database. Next, a multi-criteria decision analysis method is used to comprehensively assess the patient's long-term and short-term health trends based on this database. This step requires consideration of multiple factors, such as the stability of health trends, risk levels, and the impact of key characteristics. Finally, based on the results of the multi-criteria decision analysis, a comprehensive evaluation report for the patient is generated, including short-term risk predictions, long-term health change trends, and personalized medical intervention recommendations.

[0140] In this embodiment of the application, it is assumed that a hypertension management platform is being developed, and the patient classification results, risk outcomes, and key characteristics of health trends have been obtained through the preceding steps.

[0141] Integrating patient health information: Patient A's health trend is initially classified as "stable". Patient A's risk score for hypertensive crisis within the next 6 months is 0.2 (low risk). Key features extracted from the dimensionality-reduced data include "daily blood pressure fluctuation range", "weekly exercise frequency", and "dietary salt intake". This information is integrated into a patient health information database, as shown in Table 1 below:

[0142] Table 1

[0143] Patient ID Health Trends Classification Short-term risk score Key Feature 1 Key Feature 2 Key Feature 3 A Stablize 0.2 Daily blood pressure fluctuation range Weekly exercise frequency Dietary salt intake

[0144] The Analytic Hierarchy Process (AHP) was used to assign weights to the patient's health information. For example, the weight of the health trend classification was 0.4, the weight of the short-term risk score was 0.3, and the weight of the key features was 0.3. Each key feature was then further subdivided into scores, such as "daily blood pressure fluctuation" being scored 0.8, "weekly exercise frequency" being scored 0.6, and "dietary salt intake" being scored 0.7.

[0145] Overall score = 0.4 × Health trend classification score + 0.3 × Short-term risk score + 0.3 × (0.8 × Daily blood pressure fluctuation score + 0.6 × Weekly exercise frequency score + 0.7 × Dietary salt intake score). Assuming a health trend classification score of 0.7 and a short-term risk score of 0.2, then: Overall score = 0.4 × 0.7 + 0.3 × 0.2 + 0.3 × (0.8 × 0.8 + 0.6 × 0.6 + 0.7 × 0.7) = 0.28 + 0.06 + 0.3 × (0.64 + 0.36 + 0.49) = 0.28 + 0.06 + 0.3 × 1.49 = 0.28 + 0.06 + 0.447 = 0.787

[0146] The overall assessment indicates that Patient A has a low risk of hypertensive crisis within the next 6 months, with a score of 0.2. Based on a composite score of 0.787, Patient A's overall health trend is stable, but attention should be paid to two key characteristics: "daily blood pressure fluctuations" and "dietary salt intake," to maintain long-term health. Patient A is advised to maintain regular exercise, engaging in moderate-intensity exercise at least three times per week. Salt intake should be reduced to no more than 6 grams per day. Patient A is also advised to measure blood pressure regularly every day and record blood pressure fluctuations for timely adjustments to the treatment plan.

[0147] Figure 3 This application provides a schematic diagram of the structure of a blood gas index trend imbalance early warning and prediction system, as shown in the embodiment of the present application. Figure 3 As shown, the device includes:

[0148] The acquisition module 21 is used to collect the patient's blood gas index data, including pH value, oxygen partial pressure, carbon dioxide partial pressure and blood oxygen saturation.

[0149] The calculation module 22 is used to perform a comprehensive score on the blood gas index data using a fuzzy logic algorithm to obtain a comprehensive score result. The fuzzy logic algorithm defines fuzzy sets and membership functions and combines them with an uncertainty measure based on evidence theory to transform qualitative medical judgments into quantitative scores, assess the patient's acid-base balance and oxygenation status, and obtain a comprehensive score result.

[0150] Classification module 23 is used to classify the patient's health trend based on the comprehensive score result to obtain a classification result, wherein the classification result includes at least good, warning or critical.

[0151] Analysis module 24 is used to identify patients whose classification results include warnings or critical conditions, and to assess the short-term health risks of the patients using a survival analysis method to obtain risk results. The survival analysis method uses a Cox proportional hazards model and an Aalen multiplicative model to predict the patients' survival time and risk factors and uses these as risk results.

[0152] The determination module 25 is used to capture the long-term dependencies in the patient's historical blood gas index data using a long short-term memory network model, and to analyze the changing trends and periodicity of the blood gas index data in combination with the comprehensive score results, so as to determine the patient's preliminary health trend.

[0153] The extraction module 26 is used to perform unsupervised feature learning and dimensionality reduction on patients with similar health trends based on the patient's preliminary health trend, and extract the key features of the health trend.

[0154] The assessment module 27 is used to comprehensively evaluate the patient's target health trend based on the classification results, the risk results, and the key features of the health trend.

[0155] Figure 3 The aforementioned blood gas index trend imbalance early warning and prediction system can perform... Figure 1 The implementation principle and technical effects of the blood gas index trend imbalance early warning and prediction method described in the illustrated embodiment will not be repeated here. The specific operation methods of each module and unit in the blood gas index trend imbalance early warning and prediction system described in the above embodiments have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0156] In one possible design, Figure 3 The blood gas index trend imbalance early warning and prediction system of the embodiment shown can be implemented as a computing device, such as... Figure 4 As shown, the computing device may include a storage component 31 and a processing component 32;

[0157] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0158] The processing component 32 is used for: collecting patients' blood gas index data; using fuzzy logic algorithms to comprehensively score the blood gas index data, assessing the patients' acid-base balance and oxygenation status to obtain a comprehensive score result; obtaining a classification result based on the comprehensive score result; determining that the classification result includes patients with warning or critical conditions, obtaining a risk result, and using survival analysis methods such as the Cox proportional hazards model and the Aalen multiplicative model to predict the patients' survival time and risk factors as the risk result; analyzing the changing trends and periodicity of blood gas index data to determine the patients' preliminary health trends; performing unsupervised feature learning and dimensionality reduction on patients with similar health trends, and extracting key features of the health trends; and comprehensively evaluating the patients' target health trends.

[0159] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0160] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0161] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0162] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0163] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0164] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0165] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown is a method for early warning and prediction of blood gas index trend imbalance.

[0166] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0167] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0168] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0169] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for early warning and prediction of imbalances in blood gas indicators, characterized in that, include: Collect patients' blood gas parameters, including pH value, partial pressure of oxygen, partial pressure of carbon dioxide, and blood oxygen saturation; The blood gas index data are comprehensively scored using a fuzzy logic algorithm to obtain a comprehensive score result. The fuzzy logic algorithm defines fuzzy sets and membership functions, and combines uncertainty measures based on evidence theory to transform qualitative medical judgments into quantitative scores, assessing the patient's acid-base balance and oxygenation status to obtain a comprehensive score result. Based on the comprehensive score results, the patient's health trends are classified to obtain classification results, which include at least good, warning, or critical. The classification results are determined to include patients with warnings or critical conditions, and the short-term health risks of the patients are assessed using survival analysis methods to obtain risk results. The survival analysis methods predict the patients' survival time and risk factors using Cox proportional hazards models and Aalen multiplicative models and use these predictions as risk results. The long short-term memory network model is used to capture the long-term dependencies in the patient's historical blood gas index data, and combined with the comprehensive score results, the changing trends and periodicity of the blood gas index data are analyzed to determine the patient's preliminary health trend. Based on the patients' preliminary health trends, unsupervised feature learning and dimensionality reduction are performed on patients with similar health trends, and key features of health trends are extracted. Based on the classification results, the risk results, and the key features of the health trends, the patient's target health trends are comprehensively evaluated.

2. The method according to claim 1, characterized in that, The process of using fuzzy logic algorithm to comprehensively score the blood gas index data and obtain a comprehensive score result includes: Based on the physiological characteristics of acid-base balance and oxygenation, relevant fuzzy sets are defined, with each fuzzy set representing a specific physiological state. Using the defined fuzzy set, a membership function is established. The membership function is based on the actual measured values ​​of blood gas index data to determine the membership degree of the corresponding fuzzy set. Based on the established membership function, the blood gas index data are processed to calculate the membership degree of the fuzzy set corresponding to each blood gas index data. By utilizing the membership degree of the fuzzy set corresponding to each of the blood gas index data, combined with the uncertainty measure based on evidence theory, the uncertainty and conflict between different blood gas index data are integrated to obtain the uncertainty measure result; Based on the uncertainty measurement results, the qualitative medical judgments obtained for the patient are converted into quantitative scores, and a comprehensive score is generated based on the quantitative scores. The comprehensive score is used to assess the patient's acid-base balance and oxygenation status.

3. The method according to claim 2, characterized in that, The process of classifying the patient's health trends based on the comprehensive score results to obtain classification results includes: Based on the comprehensive scoring results, a classification standard for health trends is defined, which includes a preset scoring threshold to distinguish different categories of health trends. Multiple decision trees are constructed. Each decision tree is trained independently by randomly sampling and selecting features from the training dataset. Each decision tree performs a preliminary classification of the patient's health trend based on the comprehensive scoring results and the classification criteria. Based on the classification results of multiple decision trees, the patient's final health trend classification is determined and used as the classification result.

4. The method according to claim 3, characterized in that, The process involves identifying patients whose classification results include warnings or critical conditions, and using survival analysis to assess the short-term health risks of these patients to obtain risk outcomes, including: The Cox proportional hazards model is used to predict the survival time and risk factors of patients with warnings or critical conditions. The model expression of the Cox proportional hazards model is as follows: ;in, Indicates having covariates Individuals in time The risk function; represents the baseline risk function, and represents the risk function when there are no covariates influencing the risk. Indicates time and covariates A function representing how risk factors change over time and with individual characteristics; Indicates the first One covariate, where a covariate refers to a risk factor; Indicates time The function represents the first... and the The impact of the interaction between individual covariates on the risk function; Representing a nonlinear function, it refers to a model based on machine learning. These represent model parameters, used to capture complex nonlinear relationships between covariates; Representing covariates The index; Representing covariates The index; Indicates the total number of covariates; The Aalen multiply-add model was used to nonparametrically estimate the survival time of patients with warnings or critical conditions, yielding the instantaneous hazard ratio. The model expression for the Aalen multiply-add model is as follows: ； in, Indicates time The instantaneous risk rate; represents the baseline risk function, and represents the risk function when there are no covariates influencing the risk. Indicates time and covariates The function represents the first... The impact of each covariate on the risk function; Indicates the first Covariates over time The value of represents the change of the covariate over time; Indicates time The function represents the first... and the The impact of the interaction between individual covariates on the risk function; Representing a nonlinear function, it refers to a model based on machine learning. These represent model parameters, used to capture complex nonlinear relationships between covariates; Representing covariates The index; Representing covariates The index; Indicates the total number of covariates; Based on the prediction results of the Cox proportional hazards model and the Aalen multiplicative model, the patient's survival time and risk factors are comprehensively assessed to obtain the risk outcome.

5. The method according to claim 4, characterized in that, The method employs a long short-term memory network model to capture long-term dependencies in the patient's historical blood gas index data, and combines this with the comprehensive scoring results to analyze the changing trends and periodicity of the blood gas index data, thereby determining the patient's preliminary health trend, including: Collect the patient's historical blood gas parameters data, including time series data of pH value, partial pressure of oxygen, partial pressure of carbon dioxide, and blood oxygen saturation; A long short-term memory network model is constructed, and the gating mechanism of the long short-term memory network model is used to capture the long-term dependencies and short-term patterns in the patient's historical blood gas index data; The historical blood gas index data and the comprehensive score results are used as inputs into the long short-term memory network model to analyze the changing trends and periodicity of the blood gas index data. Based on the output of the Long Short-Term Memory Network model, the patient's preliminary health trend is determined, which includes the trend of changes in acid-base balance and the periodic changes in oxygenation.

6. The method according to claim 5, characterized in that, Based on the patient's preliminary health trend, unsupervised feature learning and dimensionality reduction are performed on patients with similar health trends, and key features of the health trends are extracted, including: Based on the patients’ preliminary health trends, a group of patients with similar health trend characteristics were selected. Select or construct an unsupervised learning model, which includes an autoencoder algorithm for learning feature representations from data corresponding to patients with similar health trends; The unsupervised learning model is used to perform feature learning and dimensionality reduction on the data corresponding to patients with similar health trends, in order to extract the latent structure and pattern in the data and map the high-dimensional data to a low-dimensional space to obtain the dimensionality-reduced data. Key features are extracted from the dimensionality-reduced data, which are used to represent the core changes and patterns in the patient's health trends.

7. The method according to claim 6, characterized in that, The comprehensive evaluation of the patient's target health trend based on the classification results, the risk results, and the key features of the health trend includes: The classification results, risk results, and key features of the health trends are integrated to form a patient health information database; Using a multi-criteria decision analysis method, based on the patient health information database, a comprehensive evaluation of the patient's long-term and short-term health trends is conducted to obtain a comprehensive evaluation result; Based on the comprehensive evaluation results, the patient's target health trend is output, including short-term risk prediction, long-term health change trend, and personalized medical intervention recommendations.

8. A blood gas index trend imbalance early warning and prediction system, characterized in that, include: The data acquisition module is used to collect patients' blood gas parameters, including pH value, partial pressure of oxygen, partial pressure of carbon dioxide, and blood oxygen saturation. The calculation module is used to perform a comprehensive score on the blood gas index data using a fuzzy logic algorithm to obtain a comprehensive score result. The fuzzy logic algorithm defines fuzzy sets and membership functions, and combines them with an uncertainty measure based on evidence theory to transform qualitative medical judgments into quantitative scores, assess the patient's acid-base balance and oxygenation status, and obtain a comprehensive score result. The classification module is used to classify the patient's health trend based on the comprehensive score result, and obtain the classification result, which includes at least good, warning or critical. An analysis module is used to identify patients whose classification results include warnings or critical conditions, and to assess the short-term health risks of the patients using survival analysis methods to obtain risk results. The survival analysis methods use a Cox proportional hazards model and an Aalen multiplicative model to predict the patients' survival time and risk factors and use these as risk results. The determination module is used to capture long-term dependencies in the patient's historical blood gas index data using a long short-term memory network model, and to analyze the changing trends and periodicity of the blood gas index data in combination with the comprehensive score results, so as to determine the patient's preliminary health trend. The extraction module is used to perform unsupervised feature learning and dimensionality reduction on patients with similar health trends based on the patient's preliminary health trends, and to extract the key features of the health trends. The assessment module is used to comprehensively evaluate the patient's target health trend based on the classification results, the risk results, and the key features of the health trend.

9. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement a blood gas index trend imbalance early warning and prediction method as described in any one of claims 1-7.

10. A computer storage medium, characterized in that, The system contains a computer program that, when executed by a computer, implements a method for early warning and prediction of blood gas index trend imbalance as described in any one of claims 1-7.

Images