Data-driven integrated management system and method for medical treatment and prevention

Abstract

The present invention relates to an integrated management system and method for medical treatment and prevention based on data application, belonging to the field of medical information technology. It includes: S1, acquiring the patient's static feature vector and dynamic temporal feature matrix. This invention can dynamically adjust according to the real-time risk of each decision scenario, possessing adaptive capabilities. It transforms the degree of conflict between the universality of guidelines and the specificity of individual data into a calculable variable, using it as a key input for dynamically adjusting the decision threshold, thus achieving deep coupling and fusion analysis between the two. Through clinical instability scoring, it transforms the dynamic evolution of the patient's condition over time into an explicit, quantifiable, and objective score, making it a formal and influential variable in the decision-making model, promoting effective human-machine collaboration.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, specifically to an integrated management system and method for medical treatment and prevention based on data applications. Background Technology

[0002] Chronic disease management is a major challenge facing the modern medical system. In order to improve management efficiency and effectiveness, an integrated medical and preventive healthcare management system has emerged. This type of system aims to provide patients with closed-loop management services covering the entire process of pre-diagnosis, diagnosis, and post-diagnosis by integrating multi-dimensional information such as patients' diagnosis and treatment information, public health records, and vital sign monitoring data.

[0003] Existing integrated medical and preventive healthcare systems typically include a built-in clinical decision support module. This module provides doctors with suggestions on treatment pathways, laboratory tests, medication recommendations, and even referral suggestions based on standardized treatment guidelines such as the "National Guidelines for the Prevention and Management of Hypertension in Primary Care." For example, the system may recommend or prohibit certain medications based on the patient's disease grade, age, and symptoms, or remind doctors to prescribe specific tests for patients on a fixed schedule such as annually or quarterly. At the same time, the system also collects individualized information about patients, such as real-time vital signs, lifestyle, and historical follow-up records, and attempts to generate personalized health prescriptions.

[0004] However, existing technologies have revealed fundamental limitations in practice; the core problem lies in the fact that when processing standardized guidelines and individualized data, systems typically employ static, rule-based, or simple weighting methods, lacking a deep, dynamic coupling mechanism between the two. This fragmented approach leads to severe technical bottlenecks when dealing with patients with complex conditions, multiple complications, or symptoms not explicitly covered by guidelines.

[0005] Specifically, when the external variable of a patient's disease state is dynamically evolving, for example, when a hypertensive patient develops acute renal failure in a short period of time, the system will face a dilemma: due to the universality and lag of guidelines, it may make inapplicable or even harmful recommendations for that specific patient, such as recommending antihypertensive drugs that may cause potential kidney damage; it may make decisions that deviate from the consensus of evidence-based medicine and have potential risks.

[0006] In this context, a sharp conflict arises between the certainty of standardized guidelines and the uncertainty of individual data. Existing technologies cannot effectively quantify the coupling effect of this conflict, leading to dynamic drift and ambiguity in the safety boundaries of decision-making. The system cannot achieve an optimal balance between ensuring safety and pursuing efficacy. As a result, when generating the optimal treatment pathway that takes into account both safety and efficacy, it faces challenges due to its inability to effectively address the conflict of decision evidence and risk uncertainty caused by individual specificity at the edge of the standardized knowledge system.

[0007] Therefore, there is an urgent need for a new technology that can dynamically quantify and address this coupling effect, and intelligently adjust the safety boundary of decision-making.

[0008] The information disclosed in the background section above is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0009] The purpose of this invention is to provide an integrated management system and method for medical and preventive medicine based on data applications, so as to solve the problems mentioned in the background art.

[0010] The technical solution of the present invention includes: S1, obtaining the patient's static feature vector and dynamic temporal feature matrix;

[0011] S2. Based on the preset clinical guideline knowledge base, combined with the static feature vector and dynamic temporal feature matrix, evaluate the clinical behavior to be decided and calculate the guideline compliance score.

[0012] S3. Based on the static feature vector and the dynamic temporal feature matrix, a machine learning model is used to predict the individualized risk of the clinical behavior to be decided and to determine the individualized risk score.

[0013] S4. Calculate the decision conflict index by combining the guideline compliance score and the individualized risk score;

[0014] S5. Calculate the clinical instability score based on the dynamic temporal feature matrix; combine the decision conflict index with the clinical instability score to generate a dynamic decision safety boundary threshold.

[0015] S6. Based on the guideline compliance score and the individualized risk score, generate a comprehensive clinical behavior recommendation score;

[0016] S7. Based on the comprehensive recommendation score of the clinical behavior and the dynamic decision safety boundary threshold, output a decision signal for recommendation or high-risk warning.

[0017] Preferably, the steps for calculating the guideline compliance score specifically include:

[0018] S21. Based on the static feature vector, the dynamic temporal feature matrix, and the preset clinical guideline knowledge base;

[0019] S22. Match the patient data with multiple rules in the knowledge base that are related to the clinical behavior to be decided, and convert the matching result of each rule into a preset score;

[0020] S23. Multiply the preset score of each rule by its corresponding medical importance weight, and sum all the products to generate the guideline compliance score.

[0021] Preferably, the steps for calculating the decision conflict index specifically include:

[0022] S41. Based on the guideline compliance score and the individualized risk score;

[0023] S42. Multiply the individualized risk score by a constant negative two and add it to a constant one to linearly map the risk probability semantics of the individualized risk score to the same recommendation and taboo semantic range as the guideline compliance score;

[0024] S43. Calculate the difference between the guideline compliance score and the mapped individualized risk score, and take the absolute value of the difference to generate the decision conflict index used to quantify the degree of contradiction between the guideline recommendations and individual risk predictions.

[0025] Preferably, the steps for calculating the clinical instability score specifically include:

[0026] S511. Diagnose complications based on the preset key indicators in the dynamic time-series feature matrix and the recently added complications;

[0027] S512. Calculate the coefficient of variation of the key indicator within the recent time window, and sum the coefficients of variation by multiplying them by their corresponding clinical importance weights to obtain the first risk component.

[0028] S513. Multiply the number of newly diagnosed complications by the preset diagnostic weighting coefficient to obtain the second risk component;

[0029] S514. Add the first risk component and the second risk component, and normalize them using the hyperbolic tangent function to generate the clinical instability score used to assess the severity of the dynamic evolution of the patient's disease state.

[0030] Preferably, the clinical importance weight and the diagnostic weight coefficient are initially set by expert scoring, and then fine-tuned by an optimization algorithm using historical case data with the goal of maximizing the correlation between the clinical instability score and actual clinical deterioration events.

[0031] Preferably, the steps for generating dynamic decision-making security boundary thresholds specifically include:

[0032] S521. Based on the decision conflict index, the clinical instability score, the preset basic safety boundary threshold, and the boundary adjustment sensitivity coefficient;

[0033] S522. Multiply the clinical instability score by the decision conflict index, and then multiply the product by the boundary adjustment sensitivity coefficient;

[0034] S523. Perform hyperbolic tangent function processing on the multiplication result to generate an adjustment factor;

[0035] S524. Multiply the adjustment factor by the difference between the adjustment factor and the base safety boundary threshold, and add the product to the base safety boundary threshold to generate the dynamic decision safety boundary threshold for adaptively adjusting the decision prudence threshold.

[0036] Preferably, the basic safety boundary threshold is preset by domain experts based on the risk preferences and medical quality requirements of medical institutions; the boundary adjustment sensitivity coefficient is optimized and determined on a dataset of labeled historical cases through cross-validation and grid search, with the goal of maximizing a preset evaluation index that balances safety and efficacy.

[0037] Preferably, the step of outputting the decision signal specifically includes:

[0038] When the comprehensive recommendation score of the clinical behavior is greater than or equal to the dynamic decision safety boundary threshold, the recommendation decision signal is output.

[0039] When the comprehensive recommendation score of the clinical behavior is less than the dynamic decision safety boundary threshold, the decision signal of the high-risk warning is output, and the guideline compliance score, individualized risk score, decision conflict index and clinical instability score are displayed as decision reference information.

[0040] A data-driven integrated management system for medical treatment and disease prevention includes:

[0041] The data acquisition and preprocessing module is used to acquire the patient's static feature vector and dynamic temporal feature matrix;

[0042] The dual-core risk assessment module is connected to the data acquisition and preprocessing module, and the dual-core risk assessment module is used for:

[0043] Based on a pre-set clinical guideline knowledge base, and combined with the static feature vector and dynamic temporal feature matrix, the clinical behavior to be decided is evaluated, and the guideline compliance score is calculated.

[0044] Based on the static feature vector and dynamic temporal feature matrix, a machine learning model is used to predict the individualized risk of the clinical behavior to be decided and to determine the individualized risk score.

[0045] A decision-making security boundary dynamic coupling module, connected to the dual-core risk assessment module, is used for:

[0046] The decision conflict index is calculated by combining the guideline compliance score and the individualized risk score.

[0047] Based on the dynamic temporal feature matrix, a clinical instability score is calculated, and a dynamic decision safety boundary threshold is generated by combining the decision conflict index with the clinical instability score.

[0048] The decision generation and recommendation module is connected to the dual-core risk assessment module and the decision security boundary dynamic coupling module. The decision generation and recommendation module is used for:

[0049] Based on the guideline compliance score and the individualized risk score, a comprehensive clinical behavior recommendation score is generated;

[0050] The decision signal is output by comparing the comprehensive recommendation score of the clinical behavior with the dynamic decision safety boundary threshold, and then outputting a recommendation decision or a high-risk warning signal.

[0051] This invention provides an improved integrated management system and method for medical treatment and prevention based on data applications, which has the following improvements and advantages compared with the prior art:

[0052] 1. This invention introduces a dynamic decision safety boundary threshold, which is driven by the decision conflict index and the clinical instability score. It can be dynamically adjusted according to the real-time risk of each decision scenario and has adaptive capability.

[0053] 2. By using the decision conflict index, the degree of conflict between the universality of the guidelines and the specificity of individual data is transformed into a calculable variable, which is then used as a key input for dynamically adjusting the decision threshold, thus achieving deep coupling and fusion analysis between the two.

[0054] 3. By using the clinical instability score, the dynamic evolution of the patient's condition over time is transformed into an explicit, quantifiable, and objective score, making it a formal and influential variable in the decision-making model. Attached Figure Description

[0055] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0056] Figure 1 This is a flowchart of Embodiment 1;

[0057] Figure 2 This is a schematic diagram illustrating the steps for calculating the guideline compliance score in Example 2 of this embodiment;

[0058] Figure 3 This is a schematic diagram of the steps for calculating the clinical instability score in Example 3 of this embodiment;

[0059] Figure 4This is a flowchart illustrating the steps involved in generating the dynamic decision-making safety boundary threshold in Example 4. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0061] Example 1:

[0062] Please see Figures 1 to 4 This invention provides a data-based integrated management system and method for medical and preventive medicine, including:

[0063] S1. Obtain the patient's static feature vector and dynamic temporal feature matrix;

[0064] S2. Based on the preset clinical guideline knowledge base, combined with static feature vectors and dynamic time-series feature matrices, evaluate the clinical behavior to be decided and calculate the guideline compliance score.

[0065] S3. Based on static feature vectors and dynamic temporal feature matrices, a machine learning model is used to predict the individualized risk of the clinical behavior to be decided and to determine the individualized risk score.

[0066] S4. Calculate the decision conflict index by combining the guideline compliance score and the individualized risk score;

[0067] S5. Calculate the clinical instability score based on the dynamic time-series feature matrix; combine the decision conflict index and the clinical instability score to generate a dynamic decision safety boundary threshold.

[0068] S6. Generate a comprehensive recommendation score for clinical behavior based on guideline compliance score and individualized risk score;

[0069] S7. Based on the comprehensive recommendation score of clinical behavior and the dynamic decision safety boundary threshold, output the decision signal of recommendation decision or high risk warning.

[0070] This embodiment provides a data-driven integrated management method for medical treatment and prevention, aiming to overcome the problem of blurred decision-making safety boundaries caused by the lack of a dynamic coupling mechanism when integrating standardized guidelines and individualized data in existing technologies. The method's technical path begins with acquiring the patient's static feature vector and dynamic temporal feature matrix, building a solid data foundation for subsequent refined assessments. The innovation of this method lies in its abandonment of static rules and the parallel initiation of a dual-core risk assessment: firstly, calculating the guideline compliance score, which quantifies the consistency between clinical behavior and the evidence-based medicine knowledge system; secondly, determining an individualized risk score through a machine learning model, proactively predicting the potential adverse event risk for specific patients. The core breakthrough of this method is that it goes beyond simply presenting these two scores side-by-side. Instead of simply counting, the system uses these two core indicators to calculate a creative decision conflict index and a clinical instability score to assess the patient's condition. These two indicators serve as key inputs, driving the generation of an unprecedented dynamic decision safety boundary threshold. A comprehensive clinical behavior recommendation score, which integrates guidelines and individualized considerations, is rigorously compared with this dynamic threshold. This mechanism enables the decision-making system to be adaptive, adjusting its prudence threshold in real time based on the degree of conflict between guidelines and individual circumstances, as well as the stability of the patient's condition. The direct effect is that the system can output clear recommended decisions or high-risk warning signals with detailed reference information, thereby building a robust and intelligent barrier between ensuring medical safety and pursuing individualized efficacy, significantly improving the quality and reliability of decision-making in complex clinical scenarios.

[0071] To enable those skilled in the art to implement this invention, a gradient boosting decision tree model, such as LightGBM or XGBoost, is preferably used to construct the machine learning model. The model construction process includes:

[0072] Patient baseline information, such as age, gender, and underlying disease diagnosis codes, is extracted from static feature vectors. From dynamic time-series feature matrices, statistical features of each key vital sign or test indicator within a preset time window are extracted, including mean, standard deviation, maximum value, minimum value, and linear regression slope, which together constitute the input features of the model.

[0073] Using a large-scale, anonymized historical electronic medical record database, the individualized risk to be predicted is defined as a specific clinical endpoint event, such as the occurrence of a major adverse cardiovascular event within 30 days of decision implementation. For each training data point, if the event occurs within the specified time, it is labeled 1; otherwise, it is labeled 0.

[0074] Using the features and labels described above, a GBDT model is trained; the model outputs a probability value between 0 and 1, which is the individualized risk score. .

[0075] Example 2:

[0076] The specific steps for calculating the guideline compliance score include:

[0077] S21. Based on static feature vectors, dynamic temporal feature matrices, and a pre-defined clinical guideline knowledge base;

[0078] S22. Match patient data with multiple rules in the knowledge base related to the clinical behavior to be decided, and convert the matching result of each rule into a preset score;

[0079] S23. Multiply the preset score of each rule by its corresponding medical importance weight, and sum all the products to generate a guideline compliance score.

[0080] The specific steps for calculating the decision conflict index include:

[0081] S41. Based on guideline compliance scores and individualized risk scores;

[0082] S42. Multiply the individualized risk score by a constant negative two and add it to a constant one to linearly map the risk probability semantics of the individualized risk score to the same recommendation and taboo semantic range as the guideline compliance score;

[0083] S43. Calculate the difference between the guideline compliance score and the mapped individualized risk score, and take the absolute value of the difference to generate a decision conflict index used to quantify the degree of contradiction between guideline recommendations and individual risk predictions.

[0084] The process of constructing a pre-defined clinical guideline knowledge base includes the following steps:

[0085] Authoritative clinical practice guidelines, such as those published by the Chinese Medical Association or the American Heart Association, are selected by a team of clinical experts, and recommendations or contraindications related to clinical behavior are manually extracted.

[0086] Each extracted rule is converted into an IF-THEN logical expression and stored in a computer-readable format; each structured rule contains at least the following fields: rule conditions, recommended behavior, rule type, and preset scores and medical importance weights, which will be explained in detail below.

[0087] In this embodiment, the step of calculating the guideline compliance score is a core step in ensuring that the decision-making is based on evidence-based medicine. The technical implementation path of this step is to perform deep matching in a preset clinical guideline knowledge base based on the obtained patient static feature vector and dynamic temporal feature matrix. For a clinical behavior to be decided, such as prescribing ACEI drugs for a hypertensive patient, the system compares the patient data with all rules related to the behavior stored in the knowledge base, such as indication rules and contraindication rules, one by one. The matching result of each rule is converted into a preset score. This score is multiplied by a pre-set medical importance weight that reflects the importance of the rule in clinical practice. Finally, the weighted scores of all rules are summed to generate the guideline compliance score.

[0088] The preset score is used to characterize the recommendation strength level of the rule itself, and its assignment refers to the general evidence level standard: for example, the preset score is 1.0 for a strong recommendation; 0.5 for a weak recommendation; -1.0 for a strong contraindication; and 0 for no clear recommendation or contraindication. The medical importance weight is a coefficient in the range [0, 1], used to reflect the severity of the clinical consequences that may result from violating the rule, and is assessed by an expert panel based on experience; for example, a rule about contraindications to avoid fatal allergic reactions should have a medical importance weight close to 1.0; while a rule about minor, manageable side effects can have a lower weight.

[0089] The step of calculating the decision conflict index constitutes the first core action of this invention in identifying and quantifying decision uncertainty. This step aims to deeply reveal the potential contradiction between standardized guidelines and individual specificity. The technical motivation stems from the fact that existing technical systems cannot effectively quantify the conflict between the universal recommendations of clinical guidelines and the specific results of individualized risk prediction. When guidelines strongly recommend a certain clinical behavior, while machine learning models predict high risk based on the patient's unique data characteristics, a severe decision conflict arises within the system. The intensity of this conflict is directly related to the risk of the decision. It is necessary to create a clear indicator to accurately capture and quantify the intensity of this conflict. The generation of this indicator is the fundamental premise and key input for initiating a more prudent decision-making mode.

[0090] Its composition is defined by the following formula:

[0091] ;

[0092] in, The decision conflict index is defined, and its value range is designed to be within a certain range. The larger the value in the range, the more severe the conflict between the guidelines and individual risk. This indicates the guideline compliance score, which is derived from the aforementioned guideline knowledge base matching and weighted summation calculation, and its value range is normalized to [value range missing]. Where 1 represents a strong recommendation and -1 represents an absolute taboo; This represents an individualized risk score, which is the predicted probability of a specific adverse event occurring, output by a machine learning model. Its value range is within... ; The functionality of this expression lies in that it sets the range to 1. Risk probability Perform a linear mapping to transform to... Completely equal semantic interval, when A result of 1 is equivalent to a strong recommendation. The result is -1, which is equivalent to a strong taboo, achieving a precise alignment between the two in terms of dimensions and medical semantics; To ensure absolute value operation, This formula measures only the magnitude of the conflict, not its direction, and is used to quantify the degree of contradiction between the guideline recommendation (A_{guide}) and the individualized risk prediction (R_{ind}).

[0093] Before each clinical behavioral decision is generated, the system performs this decision conflict index calculation; for example, if a guideline recommends a certain drug ( However, the model predicts a high risk of adverse reactions in individuals. (After mapping, it becomes -0.8), then This is a warning-indicating high conflict value; the technical effect is that it achieves accurate and dynamic quantification of the uncertainty of the core decision-making process; a high decision conflict index can serve as a strong signal to automatically trigger a highly alert state in subsequent decision-making modules of the system, and is an indispensable key input variable for realizing the dynamic adjustment mechanism of the decision safety boundary.

[0094] Example 3:

[0095] The specific steps for calculating the clinical instability score include:

[0096] S511. Diagnosis based on key indicators preset in the dynamic time series feature matrix and newly added complications;

[0097] S512. Calculate the coefficient of variation of key indicators within the recent time window, and sum the coefficients of variation by multiplying each coefficient of variation by the corresponding clinical importance weight to obtain the first risk component.

[0098] S513. Multiply the number of newly diagnosed complications in recent times by the preset diagnostic weighting coefficient to obtain the second risk component;

[0099] S514. Add the first risk component and the second risk component, and normalize them using the hyperbolic tangent function to generate a clinical instability score for assessing the severity of the dynamic evolution of the patient's disease state.

[0100] The clinical importance weight and diagnostic weight coefficients were initially set using expert scoring, and then fine-tuned using historical case data with the goal of maximizing the correlation between the clinical instability score and actual clinical deterioration events through optimization algorithms.

[0101] In this embodiment, the step of calculating the clinical instability score objectively assesses the dynamic evolution of the patient's disease state, which is a core external variable. Static decision-making models cannot capture the rapid changes in a patient's condition. A chronic disease patient who was stable yesterday may be in a highly unstable state today due to complications. This invention designs a comprehensive score to quantify the recent physiological fluctuations and the increase in the complexity of the patient's condition.

[0102] Key indicators are pre-set by clinical experts based on the specific disease; for example, for patients with heart failure, key indicators may include: vital signs such as heart rate and systolic blood pressure, laboratory test indicators such as serum sodium, serum potassium, serum creatinine, and B-type natriuretic peptide levels, as well as changes in the patient's daily weight.

[0103] The higher this score, the more unstable the patient's condition, and the more conservative the system's decision-making must be; the score is defined by the following formula:

[0104] ;

[0105] This formula assesses the severity of dynamic changes in a patient's condition; among which, The score represents clinical instability, and its range is expressed as follows: After function processing, it is close to The higher the value, the more drastic the recent fluctuations in the patient's condition and the more unstable the state. This indicates the total number of key vital signs or laboratory test indicators that are preset and need to be monitored. Indicates the first A time-series data set comprising key indicators within the most recent preset time window, defined for ease of calculation. For this time series data set The arithmetic mean within the time window; Indicates the first The coefficient of variation of an indicator is a dimensionless, standardized statistic used to measure the relative dispersion of time series data and can effectively reflect the true volatility of the indicator. Indicates the first The clinical importance weights corresponding to the key indicators reflect the differences in the contribution of fluctuations in different indicators to the assessment of overall disease stability. This indicates the number of newly diagnosed complications that potentially interact with the current primary disease within a recent time window; The weighting coefficient for newly diagnosed cases is used to quantify the impact of newly developed complications on overall instability. The hyperbolic tangent function smoothly maps the internal weighted sum to the normalized interval, thus improving the mathematical properties of the score and effectively preventing score explosions caused by drastic fluctuations in certain indicators or sudden increases in the number of complications. This enhances the numerical stability of the calculation and avoids the influence of indicator mean values. The coefficient of variation is close to 0, which leads to meaningless maxima. Therefore, the calculation of the coefficient of variation is optimized. The expression for the coefficient of variation used here is:

[0106] ;

[0107] Where Var(Vj) is a set variance Its mean, It is a preset, extremely small positive number to ensure that the denominator is not zero or too small;

[0108] To ensure that those skilled in the art can implement it, clinical importance weighting Diagnostic weighting coefficient The determination of the weighting parameters adopts a two-stage method to ensure its scientific validity and effectiveness. In the initial stage, multiple senior clinical experts assign preliminary values ​​based on their professional knowledge and clinical experience through expert scoring. Then, in the second stage, a massive, anonymous historical case database is used to optimize the correlation between the clinical instability score calculated in this embodiment and the actual clinical deterioration events recorded in the historical data. The optimization algorithm, such as genetic algorithm or gradient optimization, is used to finely and automatically adjust these initially set weighting parameters.

[0109] Example 4:

[0110] The specific steps for generating dynamic decision-making safety boundary thresholds include:

[0111] S521, based on decision conflict index, clinical instability score, preset basic safety boundary threshold and boundary adjustment sensitivity coefficient;

[0112] S522. Multiply the clinical instability score by the decision conflict index, and then multiply the product by the boundary adjustment sensitivity coefficient.

[0113] S523. Perform hyperbolic tangent function processing on the multiplication result to generate an adjustment factor;

[0114] S524. Multiply the adjustment factor by the difference between the adjustment factor and the base safety boundary threshold, and add the product to the base safety boundary threshold to generate a dynamic decision safety boundary threshold for adaptively adjusting the decision prudence threshold.

[0115] The basic safety boundary threshold is preset by domain experts based on the risk preferences and medical quality requirements of medical institutions; the boundary adjustment sensitivity coefficient is optimized and determined on a dataset of labeled historical cases through cross-validation and grid search, with the goal of maximizing the preset evaluation indicators that balance safety and efficacy.

[0116] In this embodiment, the step of generating a dynamic decision safety boundary threshold is the key to the technical solution of this invention. Using the aforementioned decision conflict index and clinical instability score as input, a final threshold that can adaptively adjust the decision prudence threshold is generated. Any static and unchanging decision threshold cannot meet the requirements in a complex and ever-changing clinical environment. The risk of decision-making is not constant and will rise sharply as the conflict between guidelines and individual circumstances intensifies and the patient's condition deteriorates. The decision-making system must have self-restraint capabilities, that is, actively raise the threshold for automatic adoption of decisions and force decision recommendations to have higher certainty and safety. This dynamic safety boundary model is designed to realize such an intelligent adjustment mechanism that dynamically tightens or relaxes the decision threshold according to the risk.

[0117] The composition is defined by the following formula:

[0118] ;

[0119] in, This represents the dynamic decision safety boundary threshold of the final output, and its value range is limited to... The interval represents the minimum score required for a decision to pass; This represents the basic security boundary threshold, which is a basic parameter that can be preset according to specific application scenarios. It represents the default level of decision-making prudence under the most ideal conditions. This represents the clinical instability score calculated in the previous step; This represents the decision conflict index calculated in the previous step; This represents the boundary adjustment sensitivity coefficient, a crucial adjustable hyperparameter that directly determines the safety boundary. The rate and magnitude of increase with the growth of conflict and instability; This factor determines the maximum theoretical space in which the threshold can be raised; The hyperbolic tangent function, when used here, ensures that the threshold growth process is smooth and bounded, effectively preventing meaningless divergent growth of the threshold due to extreme input values.

[0120] In application, when the patient's condition is stable ( And the guidelines are consistent with individual risk. )hour, The system uses a standard threshold; when the patient's condition deteriorates rapidly ( Furthermore, the guidelines are in serious conflict with individual risk. )hour, The threshold for decision-making will approach 1, which significantly increases the threshold for decision-making. The technical effect of this model is that it achieves adaptive adjustment of decision-making prudence, enabling the system to operate efficiently in simple scenarios and automatically switch to a highly secure conservative mode in complex scenarios.

[0121] To ensure feasibility, basic safety boundary thresholds The settings are authorized to domain experts or medical institution managers, who can pre-set them based on the institution's risk preferences and medical quality requirements; while the boundary adjustment sensitivity coefficient The determination process is a rigorous data-driven optimization process. This process is conducted on a large-scale, high-quality historical case dataset that has been annotated by experts. By employing cross-validation and grid search techniques, the ultimate optimization objective is to maximize a pre-defined composite evaluation index that comprehensively considers the safety and effectiveness of decision-making, thereby accurately determining the... The optimal value.

[0122] The specific steps for outputting the decision signal include:

[0123] When the comprehensive recommendation score of clinical behavior is greater than or equal to the dynamic decision safety boundary threshold, a recommendation decision signal is output.

[0124] When the overall clinical behavior recommendation score is less than the dynamic decision safety boundary threshold, a high-risk warning decision signal is output, and the guideline compliance score, individualized risk score, decision conflict index, and clinical instability score are displayed as decision reference information.

[0125] In this embodiment, the step of outputting the decision signal is the endpoint of the entire methodology. Based on all preceding calculations and evaluations, a clear and actionable final conclusion is generated. The system first calculates a comprehensive clinical behavior recommendation score. This score combines the guidance of the guidelines with the uniqueness of individual risk, and is calculated as follows:

[0126] ;

[0127] Weight and It is not fixed, but rather optimized and determined through a data-driven approach. On a labeled, independent validation dataset, with the goal of maximizing a pre-defined comprehensive clinical utility indicator, such as the F1-score or Matthews correlation coefficient that balances recommendation accuracy and high-risk avoidance rate, grid search and other methods are used to find the optimal value within the interval [0, 1]. Numerical values; this process ensures that the fusion weighting of guideline evidence and individualized evidence achieves optimal results in specific clinical scenarios;

[0128] in, and These represent the weights of the guideline compliance score and the semantically mapped individualized risk score, respectively. ;

[0129] The subsequent decision-making logic is clear and binary;

[0130] When the calculated comprehensive recommendation score for clinical behavior Greater than or equal to the dynamic decision safety boundary threshold generated in the preceding steps When this occurs, it indicates that the overall recommendation strength of the clinical behavior has successfully crossed the safety threshold dynamically set by the system based on the current risk level. The system will then directly output a recommendation decision signal to explicitly suggest that clinicians adopt the behavior.

[0131] Conversely, when the overall recommended score for clinical behavior is... Failed to reach the dynamic decision-making safety boundary threshold If the system determines that the behavior carries a significant risk or uncertainty, it will output a high-risk warning signal. This warning signal is not a simple negative conclusion, but is accompanied by a wealth of decision-making reference information. The system will incorporate the core quantitative indicator of the entire decision-making process, namely the guideline compliance score. Individualized risk score Decision-making conflict index and clinical instability score This information is presented to clinicians in a structured manner; this design greatly enhances the system's transparency and interpretability. It not only warns of risks but also reveals their sources in greater depth, thereby empowering clinicians to make final human decisions with unprecedented information support, achieving efficient human-machine collaboration.

[0132] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A medical and health prevention integration management method based on data application, characterized in that, The method comprises the following steps: S1, obtaining a static feature vector and a dynamic time sequence feature matrix of a patient; S2, based on a preset clinical guideline knowledge base, combining the static feature vector and the dynamic time sequence feature matrix, evaluating a to-be-decided clinical behavior, and calculating a guideline compliance score; S3, based on the static feature vector and the dynamic time sequence feature matrix, predicting an individualized risk of the to-be-decided clinical behavior through a machine learning model, and determining an individualized risk score; S4, combining the guideline compliance score and the individualized risk score, calculating a decision conflict index; S5, based on the dynamic time sequence feature matrix, calculating a clinical instability score; combining the decision conflict index and the clinical instability score, and generating a dynamic decision safety boundary threshold; S6, based on the guideline compliance score and the individualized risk score, generating a comprehensive recommendation score of the clinical behavior; S7, comparing the comprehensive recommendation score of the clinical behavior with the dynamic decision safety boundary threshold, and outputting a recommended decision or a decision signal of high-risk warning; The step of calculating the guideline compliance score comprises the following steps: S21, based on the static feature vector, the dynamic time sequence feature matrix, and the preset clinical guideline knowledge base; S22, matching the patient data with a plurality of rules related to the to-be-decided clinical behavior in the knowledge base, and converting the matching result of each rule into a preset score; S23, multiplying the preset score of each rule by its corresponding medical importance weight, and summing all the products to generate the guideline compliance score; The step of calculating the decision conflict index comprises the following steps: S41, based on the guideline compliance score and the individualized risk score; S42, multiplying the individualized risk score by a constant negative two and adding a constant one to linearly map the risk probability semantics of the individualized risk score to the same recommendation and taboo semantic interval as the guideline compliance score; S43, calculating the difference between the guideline compliance score and the mapped individualized risk score, and taking the absolute value of the difference to generate the decision conflict index for quantifying the degree of contradiction between the guideline recommendation and the individualized risk prediction; The constitution is defined by the following formula: where, represents the decision conflict index, whose value domain is designed in interval, the larger the value, the more intense the conflict between the guideline recommendation and the individual risk; represents the guideline adherence score, which is derived from the aforementioned guideline knowledge base matching and weighted summation calculation, whose value domain is normalized to , where 1 represents strong recommendation, and -1 represents absolute contraindication; represents the individualized risk score, which is the prediction probability of the occurrence of a specific adverse event output by the machine learning model, whose value domain is in ; The functionality of this expression is that it linearly maps the risk probability with a value domain of to the semantic interval that is completely equivalent to , when , the result is 1, equivalent to strong recommendation, and when , the result is -1, equivalent to strong contraindication, achieving the exact alignment of the two in dimension and medical semantics; is the absolute value operation, which ensures that only measures the magnitude of the conflict, not its direction, and this formula is used to quantify the degree of contradiction between the guideline recommendation (A_{guide}) and the individualized risk prediction (R_{ind}). 2.The data application-based medical and health integration management method according to claim 1, characterized in that, The step of calculating the clinical instability score comprises the following steps: S511, based on the preset key indicators in the dynamic time sequence feature matrix and the newly added complication diagnosis in the recent period; S512, calculating the coefficient of variation of the key indicators in the recent time window, multiplying each coefficient of variation by the corresponding clinical importance weight, and summing to obtain a first risk component; S513, multiplying the number of newly added complication diagnoses by a preset diagnosis weight coefficient to obtain a second risk component; S514, adding the first risk component and the second risk component, and performing normalization processing through a hyperbolic tangent function to generate the clinical instability score for evaluating the dynamic evolution degree of the patient's disease state. 3.The medical and health prevention fusion integration management method based on data application according to claim 2, characterized in that, The clinical importance weight and the diagnostic weight coefficient are preliminarily set by expert scoring method, and then are fine-tuned by optimization algorithm by using historical case data to maximize the correlation between the clinical instability score and the real occurrence of clinical deterioration events. 4.The data application-based medical and health integration management method according to claim 1, characterized in that, The step of generating the dynamic decision safety boundary threshold value specifically comprises: S521, based on the decision conflict index, the clinical instability score, a preset basic safety boundary threshold value and a boundary adjustment sensitivity coefficient; S522, multiplying the clinical instability score and the decision conflict index, and then multiplying the product and the boundary adjustment sensitivity coefficient; S523, performing hyperbolic tangent function processing on the multiplication result to generate an adjustment factor; S524, multiplying the adjustment factor and a difference value obtained by subtracting the basic safety boundary threshold value from the basic safety boundary threshold value, and adding the product and the basic safety boundary threshold value to generate the dynamic decision safety boundary threshold value for adaptively adjusting the decision prudence threshold; The basic safety boundary threshold value is preset by domain experts according to the risk preference and medical quality requirement of a medical institution; and the boundary adjustment sensitivity coefficient is optimized and determined by cross-validation and grid search on a labeled historical case data set to maximize a preset evaluation index considering safety and efficacy. ； wherein, represents the dynamic decision safety boundary threshold of the final output, whose value range is limited in interval, representing the minimum score requirement for decision passing; represents the basic safety boundary threshold, which is a basic parameter that can be preset according to specific application scenarios, and it represents the default decision prudence in the most ideal case; represents the clinical instability score calculated in the previous step; represents the decision conflict index calculated in the previous step; represents the boundary adjustment sensitivity coefficient, which is a crucial controllable hyperparameter, directly determining the rate and amplitude of the safety boundary increase with the increase of conflict and instability; This item determines the maximum theoretical space for the threshold to float up; is a hyperbolic tangent function, its application here ensures that the growth process of the threshold is smooth and bounded, which can effectively prevent the threshold from appearing meaningless divergent growth due to extreme input values.

5. The medical and health management method based on data application integration according to claim 4, characterized in that, The step of outputting the decision signal specifically comprises: 6.The data application-based medical and health prevention fusion integration management method according to claim 1, characterized in that, When the clinical behavior comprehensive recommendation score is greater than or equal to the dynamic decision safety boundary threshold value, outputting the recommended decision signal; When the clinical behavior comprehensive recommendation score is less than the dynamic decision safety boundary threshold value, outputting the high-risk early warning decision signal, and displaying the guideline adherence score, the individualized risk score, the decision conflict index and the clinical instability score as decision reference information. The data acquisition and preprocessing module is configured to acquire a static feature vector and a dynamic time series feature matrix of a patient.

7. A medical and health care fusion integrated management system based on data application, applying the medical and health care fusion integrated management method based on data application according to any one of claims 1-6. The dual-core risk assessment module is connected with the data acquisition and preprocessing module, and is configured to: based on a preset clinical guideline knowledge base, in combination with the static feature vector and the dynamic time series feature matrix, assess a to-be-decided clinical behavior to obtain a guideline adherence score; based on the static feature vector and the dynamic time series feature matrix, predict an individualized risk of the to-be-decided clinical behavior by a machine learning model to determine an individualized risk score; The decision safety boundary dynamic coupling module is connected with the dual-core risk assessment module, and is configured to: in combination with the guideline adherence score and the individualized risk score, calculate a decision conflict index; based on the dynamic time series feature matrix, calculate a clinical instability score, and in combination with the decision conflict index and the clinical instability score, generate a dynamic decision safety boundary threshold value; The decision generation and recommendation module is connected with the dual-core risk assessment module and the decision safety boundary dynamic coupling module, and is configured to: based on the guideline adherence score and the individualized risk score, generate a clinical behavior comprehensive recommendation score; ​ ​ According to the comparison between the clinical behavior comprehensive recommendation score and the dynamic decision safety boundary threshold, a recommended decision or a high-risk early warning decision signal is output.

Images