Big data and ai-based clinical pathway deviation risk management and control tracing method, device, equipment and medium

Abstract

The application discloses a clinical path deviation risk management and control tracing method and device based on big data and AI, equipment and a medium, comprising: preprocessing multi-source heterogeneous clinical data; dividing a time sequence stage based on the preprocessed multi-source heterogeneous clinical data, and constructing a stage-specific risk feature library; constructing a time sequence causal forest model based on the stage-specific risk feature library, and combining a real-time risk cross-stage prediction mechanism to construct a double-level clinical path deviation risk early warning model; constructing a full-link causal tracing graph based on a structural causal model, and performing hierarchical positioning and tracing of the main cause of deviation; and constructing a personalized correction scheme generation model based on the results of the double-level clinical path deviation risk early warning model, the full-link causal tracing graph and the hierarchical positioning and tracing. The application belongs to the field of medical risk prediction. The application can ensure the continuity of the treatment decision scheme.

Description

Technical Field

[0001] This invention relates to the field of medical risk prediction, and in particular to a method, apparatus, equipment, and medium for tracing and managing clinical pathway deviation risks based on big data and AI. Background Technology

[0002] With the development of medical informatization, hospitals have accumulated a large amount of heterogeneous clinical data from multiple sources during the diagnosis and treatment process, including electronic medical records, laboratory test results, and medical orders. Clinical pathways, as an important tool for standardizing diagnosis and treatment behavior and improving medical quality and efficiency, have been widely used. However, in actual implementation, due to factors such as individual patient differences, differences in physician experience, and unexpected events, pathway deviations are quite common, which can easily lead to increased medical risks and decreased resource utilization efficiency.

[0003] Existing technologies mostly employ rule-based or statistical analysis methods for post-hoc evaluation or simple early warning of pathway execution, which generally suffer from insufficient real-time performance, limited predictive capabilities, and difficulty in characterizing complex causal relationships. Furthermore, traditional methods tend to focus on single-stage or single-factor analysis, lacking the ability to dynamically model the entire diagnosis and treatment process, making it difficult to systematically characterize the evolution of risk across stages. Therefore, how to construct a clinical pathway deviation risk management method that integrates multi-source data and possesses time-series modeling and causal inference capabilities to achieve real-time early warning and cross-stage prediction of risk is an urgent problem to be solved. Summary of the Invention This invention provides a method, device, equipment, and medium for tracing and managing clinical pathway deviation risks based on big data and AI, which solves the technical problem of difficulty in predicting medical risks across stages in the prior art and achieves the technical effect of predicting medical risks across stages.

[0004] Firstly, this invention provides a method for tracing and managing the source of clinical pathway deviation risks based on big data and AI, including: Acquire multi-source heterogeneous clinical data and preprocess the multi-source heterogeneous clinical data, including time period alignment and data standardization; Based on the preprocessed multi-source heterogeneous clinical data, time-series stages were divided, and a stage-specific risk feature library was constructed. Based on a stage-specific risk feature library, a time-series causal forest model is constructed, and combined with a real-time risk cross-stage prediction mechanism, a two-level clinical pathway deviation risk early warning model is built. A full-link causal source map is constructed based on the structural causal model, and the deviation from the main cause is hierarchically located and traced. Based on the two-level clinical pathway deviation risk warning model, the full-link causal traceability map, and the results of hierarchical positioning and tracing, a personalized correction plan generation model is constructed, which is used to generate clinical correction plans.

[0005] Furthermore, based on the preprocessed multi-source heterogeneous clinical data, time-series stages were divided, and a stage-specific risk feature library was constructed, including: Dividing the diagnosis and treatment stages ,in, This is a collection of diagnostic and treatment stages. For the first Each stage of diagnosis and treatment; Construct feature sets corresponding to each stage of diagnosis and treatment, including standard pathway compliance features, patient physiological state features, and clinical event risk features; Determine the weight of each feature, including:

[0006] in, For the stage The Middle The weights of each feature, Information gain ratio For the stage The Middle One characteristic, Labels for path deviation; Construct a risk transmission matrix across treatment stages, where the dimensions of the risk transmission matrix are: Dimensions include:

[0007] in, For the first Each stage of diagnosis and treatment and the first Risk transmission matrix for each stage of diagnosis and treatment For conditional probability, For the first The path deviates from the label at each stage of diagnosis and treatment. For the first The path deviates from the label at each stage of diagnosis and treatment.

[0008] Furthermore, based on a stage-specific risk feature library, a time-series causal forest model is constructed, and combined with a real-time cross-stage risk prediction mechanism, a two-tiered clinical pathway deviation risk early warning model is built, including: Based on preprocessed multi-source heterogeneous clinical data, a causal inference standard sample set is constructed, which includes sample units, label definitions, and causal processing variables. Based on the causal inference standard sample set and the stage-specific risk feature library, a time-series causal forest model is constructed, including:

[0009] in, For the first Characteristics in each stage of diagnosis and treatment The corresponding causal effect value, For the expectation, The path deviation label represents the result after the deviation has occurred. The result of labeling path deviation as path compliance For the stage Features; Based on the causal splitting criterion, a splitting gain is applied to the temporal causal forest model, including:

[0010] in, For split gain, The CATE value of the parent node. The CATE value of the left child node after the split. This is the CATE value of the right child node after the split; Determine the path deviation from the real-time risk value at the current stage of diagnosis and treatment, including:

[0011] in, For the stage The path deviates from the real-time risk value. For the stage The Middle The weights of each feature, For the stage The number of features in the middle, For the stage The Middle The causal effect value of each feature; Determine the path deviation from the real-time risk value across stages, including:

[0012] in, For the stage The path deviates from the real-time risk value. as well as To preset weights, The time decay coefficient, For the first Each stage of diagnosis and treatment and the first Risk transmission matrix for each stage of diagnosis and treatment.

[0013] Furthermore, a full-link causal source map is constructed based on a structural causal model, and the deviations from the main cause are hierarchically located and traced, including: Construct a structural causal model, which includes decision nodes, action nodes, event nodes, and result nodes; Based on the structural causal model, a full-link causal source map is constructed. ,in, For a full-link causal traceability map, For a set of nodes, It is a set of causal edges; Based on the end-to-end causal attribution map, the causal effects corresponding to each node are determined, including:

[0014] in, For nodes The causal effect value, For the node To the deviation result All valid causal pathways, The weight of each causal edge in the path. For an effective causal path, It is a causal side.

[0015] Furthermore, based on the two-tiered clinical pathway deviation risk warning model, the full-link causal tracing map, and the results of hierarchical localization and tracing, a personalized correction plan generation model is constructed, including: Based on the causal effect values ​​of each node, the path deviation is classified into levels, including fully reversible, partially reversible, and irreversible. A deep reinforcement learning model based on a deep Q-network is constructed, and the deep reinforcement learning model is used as a model for generating personalized correction schemes. The deep reinforcement learning model includes an agent, environment, state space, and action space. The reward function for constructing the personalized correction scheme generation model includes:

[0016] in, As a reward value, All are weights. For compliance rewards, As a reward for patient safety, Incentives for improving treatment efficiency As a reward for medical costs; Personalized correction schemes are generated based on a personalized correction scheme generation model.

[0017] Furthermore, the multi-source heterogeneous clinical data undergoes preprocessing, including: Timeline alignment of high-frequency continuous time-series data, discrete clinical event data, and static baseline data; Based on Z-score standardization, data standardization is performed on multi-source heterogeneous clinical data.

[0018] Also includes: The personalized correction scheme generation model is optimized based on 10-fold cross-validation.

[0019] Secondly, this invention provides a clinical pathway deviation risk management and tracing device based on big data and AI, comprising: The acquisition module is used to acquire multi-source heterogeneous clinical data and preprocess the multi-source heterogeneous clinical data, including time period alignment and data standardization. The risk feature library construction module is used to divide time-series phases based on preprocessed multi-source heterogeneous clinical data and construct phase-specific risk feature libraries. The risk warning model construction module is used to build a time-series causal forest model based on a stage-specific risk feature library, and to build a two-level clinical pathway deviation risk warning model by combining a real-time risk cross-stage prediction mechanism. The graph construction module is used to construct a full-link causal source map based on the structural causal model, and to perform hierarchical location and tracing of deviations from the main cause; The scheme generation module is used to construct a personalized correction scheme generation model based on the two-level clinical pathway deviation risk warning model, the full-link causal traceability map, and the results of hierarchical positioning and tracing. The personalized correction scheme generation model is used to generate clinical correction schemes.

[0020] Thirdly, the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to execute a clinical pathway deviation risk management and tracing method based on big data and AI, as provided in the first aspect.

[0021] Fourthly, the present invention provides a non-transitory computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor of an electronic device, enables the electronic device to execute the clinical pathway deviation risk management and tracing method provided in the first aspect, based on big data and AI.

[0022] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention achieves unified modeling and efficient utilization of information throughout the entire diagnosis and treatment process by integrating multi-source heterogeneous clinical data and performing standardized processing, significantly improving the completeness and consistency of data utilization. Through temporal stage segmentation and the construction of stage-specific risk feature databases, key risk factors at different stages of diagnosis and treatment are accurately characterized, thereby enhancing the model's ability to identify pathway deviation risks. Furthermore, the introduction of a temporal causal forest model combined with a cross-stage risk prediction mechanism enables real-time assessment and prospective prediction of clinical pathway deviation risks, improving the timeliness and accuracy of risk warnings.

[0023] This invention constructs a full-link causal source map based on a structural causal model, enabling multi-level localization and tracing of key causes of path deviations, thus enhancing the interpretability of the analysis results. By combining two-level risk warning results with causal source information, a personalized correction plan generation model is constructed, providing targeted intervention strategies for different patients. This effectively reduces the incidence of path deviations, improves medical quality and safety, and optimizes the efficiency of medical resource allocation. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A flowchart illustrating the clinical pathway deviation risk management and tracing method based on big data and AI provided by this invention; Figure 2 This is a flowchart illustrating the process of the clinical pathway deviation risk management and tracing device based on big data and AI provided by the present invention. Detailed Implementation

[0026] This invention solves the technical problems of existing clinical pathway deviation risk management and tracing methods based on big data and AI by providing a clinical pathway deviation risk management and tracing method based on big data and AI.

[0027] The technical solution of this invention is to solve the above-mentioned technical problems, and the overall idea is as follows: A big data and AI-based approach to clinical pathway deviation risk management and tracing includes: acquiring multi-source heterogeneous clinical data and preprocessing it, including time period alignment and data standardization; dividing the preprocessed multi-source heterogeneous clinical data into time-series stages and constructing a stage-specific risk feature library; constructing a time-series causal forest model based on the stage-specific risk feature library and combining it with a real-time risk cross-stage prediction mechanism to construct a two-level clinical pathway deviation risk warning model; constructing a full-link causal tracing graph based on the structural causal model and performing hierarchical location and tracing of the main causes of deviation; and constructing a personalized correction plan generation model based on the results of the two-level clinical pathway deviation risk warning model, the full-link causal tracing graph, and hierarchical location and tracing, whereby the personalized correction plan generation model is used to generate clinical correction plans.

[0028] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0029] First, it should be clarified that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0030] This invention provides, for example Figure 1 The clinical pathway deviation risk management and tracing method shown includes steps S11-S15: Step S11: Obtain multi-source heterogeneous clinical data and preprocess the multi-source heterogeneous clinical data, including time period alignment and data standardization.

[0031] Preprocessing of multi-source heterogeneous clinical data includes: timeline alignment of high-frequency continuous time-series data, discrete clinical event data, and static baseline data; and data standardization of multi-source heterogeneous clinical data based on Z-score standardization.

[0032] Specifically, multi-source heterogeneous clinical data can be divided into three categories: Continuous Time Series Data The system collects real-time patient vital signs data, including heart rate, blood pressure, blood oxygen saturation, respiratory rate, and sleep quality score, through smart wearable devices and monitoring equipment. The sampling frequency is... Simultaneously collect medication administration and symptom feedback data reported by the mobile application terminal; Discrete clinical event data The hospital information system synchronizes electronic medical records, medical orders, medications, examinations and tests, nursing procedures, clinical pathway execution records, and medical behavior rationality warning information. Each event corresponds to a unique timestamp. ; Static baseline data Patient demographic information, past medical history, admission diagnosis, standard clinical pathway protocol for enrollment, and DRG payment standard data are baseline characteristics that do not change over time.

[0033] Using standard diagnosis and treatment nodes of single-disease clinical pathways as anchor points, a globally unified timeline is constructed. ,in The starting point for patient enrollment in the clinical pathway. For the completion / discharge time of the pathway, each Standard diagnostic and treatment nodes corresponding to the pathway (such as preoperative assessment, surgical procedure, postoperative day 1, efficacy evaluation, etc.) serve as a unified benchmark for aligning the time sequence of all data.

[0034] For high-frequency continuous time-series data: For each time anchor point Define clinically meaningful alignment windows The window duration is set according to the clinical attributes of the diagnosis and treatment nodes (e.g., the preoperative assessment window is 24 hours before surgery, and the postoperative critical care monitoring window is 6 hours after surgery). High-frequency time-series data within the window Extract the time-series aggregated feature vector:

[0035] in, This represents the mean of the data within the window. The standard deviation of the data within the window. The linear fitting slope of the data within the window (characterizing the changing trend of physiological indicators) enables time-series compression of high-frequency data and extraction of clinical features, avoiding data dimensionality explosion and information loss.

[0036] For discrete clinical event data: For each clinical event (Time of occurrence) ), calculate its relationship with the nearest time anchor point Time distance:

[0037] The event Mapped to Minimum time anchor The generated event feature vector is quantized using one-hot encoding and clinical weight assignment. Clinical experts label the impact of events on the pathway (e.g., surgical procedure events have a weight of 1.0, and routine vital sign monitoring has a weight of 0.2) to ensure that the clinical impact of events can be quantified.

[0038] The aligned continuous features are standardized using Z-score to eliminate dimensional differences, as shown in the following formula:

[0039] In the formula: The mean of this feature for the training set, The standard deviation of this feature in the training set; Outlier removal and missing value imputation are completed simultaneously. Missing values ​​are imputed using a clinical rule constraint method that combines forward temporal imputation with matching imputation from patients with the same baseline. When the missing value of a key indicator exceeds the threshold, matching values ​​from patients with the same path, diagnosis, and baseline are used to imput the missing value, thus avoiding clinical logic errors caused by simple imputation.

[0040] The preprocessed full data is stored in separate databases according to the clinical pathway diagnosis and treatment stages. High-frequency data is updated incrementally every 15 minutes, and clinical event data is updated in real time. A daily full backup frequency is set, which complies with relevant medical data security and privacy protection regulations.

[0041] Step S12: Based on the preprocessed multi-source heterogeneous clinical data, time-series stages are divided, and a stage-specific risk feature library is constructed.

[0042] Specifically, it includes: Dividing the diagnosis and treatment stages ,in, This is a collection of diagnostic and treatment stages. For the first Each stage of diagnosis and treatment.

[0043] Based on the single-disease clinical pathway management guidelines, the enrolled single-disease clinical pathways were divided into: A continuous treatment phase Each stage This corresponds to a clearly defined scope of diagnosis and treatment, standard operating procedures, expected treatment goals, and time windows.

[0044] For example, the clinical pathway for community-acquired pneumonia can be divided into four core stages: admission assessment, initial anti-infection treatment, efficacy assessment, and discharge preparation.

[0045] Construct feature sets corresponding to each stage of diagnosis and treatment, including standard pathway compliance features, patient physiological state features, and clinical event risk features.

[0046] For each stage of diagnosis and treatment Three feature sets were constructed, and redundant features and noisy features with no clinical significance were removed: Standard path compliance features This stage requires the completion of the following diagnostic and treatment procedures, examinations and tests, medication guidelines, and pathway requirements, with a characteristic value of 1 (completed according to standards) or 0 (not completed / does not meet standards). Patient physiological characteristics The time-series aggregation features of physiological signs after alignment at this stage; Clinical event risk characteristics The clinical event characteristics aligned at this stage are simultaneously incorporated into evidence-based medicine features from continuing education data, clinical research data, and real-world research data.

[0047] The total set of features for each stage is: The feature dimension is .

[0048] Determine the weight of each feature, including:

[0049] in, For the stage The Middle The weights of each feature Information gain ratio For the stage The Middle One characteristic, Labels for path deviation; Construct a risk transmission matrix across treatment stages, where the dimensions of the risk transmission matrix are: Dimensions include:

[0050] in, For the first Each stage of diagnosis and treatment and the first Risk transmission matrix for each stage of diagnosis and treatment For conditional probability, For the first The path deviates from the label at each stage of diagnosis and treatment. For the first The path deviates from the label at each stage of diagnosis and treatment.

[0051] Step S13: Based on the stage-specific risk feature library, construct a time-series causal forest model, and combine it with a real-time risk cross-stage prediction mechanism to construct a two-level clinical pathway deviation risk early warning model.

[0052] Specifically, it includes: Based on preprocessed multi-source heterogeneous clinical data, a causal inference standard sample set is constructed, which includes sample units, label definitions, and causal processing variables.

[0053] Sample Unit: Full-cycle time-series data of a patient's single disease course clinical pathway, with a single sample corresponding to the complete pathway diagnosis and treatment data of a single patient in a single hospitalization cycle; Tag definition: Result label To determine whether a clinical pathway deviation has occurred, the definition of deviation must conform to the national clinical pathway management guidelines. This includes failure to complete pathway nodes according to standards, exceeding the standards for diagnosis and treatment, exceeding the DRG / pathway standards for length of stay / medical costs, and the occurrence of unexpected complications. For deviation to occur, For path compliance; Causal treatment variables: Define treatment variables For "whether or not to implement standard clinical pathway procedures", This is for cases where standard actions (processing groups) were not performed. To be performed according to the standard (control group); In addition, sample preprocessing may also be included: Sample preprocessing: Propensity score matching (PSM) was used to match the treatment group and the control group 1:1 to eliminate the influence of confounding variables such as demographic characteristics and baseline disease status, and to ensure the accuracy of causal inference. The matched sample set was divided into training set, validation set and test set in a ratio of 7:2:1.

[0054] Temporal causal forests build upon generalized random forests by incorporating strict temporal stage constraints. Their core objective is to estimate the conditional average treatment effect (CATE) of each feature on path deviation, i.e., the causal influence of that feature on path deviation.

[0055] Based on the causal inference standard sample set and the stage-specific risk feature library, a time-series causal forest model is constructed, including:

[0056] in, For the first Characteristics in each stage of diagnosis and treatment The corresponding causal effect value, For the expectation, The path deviation label represents the result after the deviation has occurred. The result of labeling path deviation as path compliance For the stage Its characteristics. The larger the positive value, the stronger the causal effect of this feature on path deviation.

[0057] The decision tree splitting criterion adopts the causal splitting criterion, which is different from the variance minimization criterion of traditional random forests, ensuring that the causal effect estimation accuracy of the child nodes after splitting is optimal.

[0058] The causal splitting criterion applies a splitting gain to the temporal causal forest model, including:

[0059] in, For split gain, The CATE value of the parent node. The CATE value of the left child node after the split. This is the CATE value of the right child node after the split.

[0060] In addition, timing constraints can be added simultaneously, namely: stage The tree model can only use arrive Historical feature data should be used, and features from future treatment stages should not be used to ensure that the model conforms to the time logic of clinical diagnosis and treatment, avoid data leakage, and guarantee clinical effectiveness.

[0061] Based on the trained temporal causal forest model, a two-tier early warning system, which is impossible to achieve with existing technologies, is constructed, including: Determine the path deviation from the real-time risk value at the current stage of diagnosis and treatment, including:

[0062] in, For the stage The path deviates from the real-time risk value. For the stage The Middle The weights of each feature For the stage The number of features, For the stage The Middle The causal effect value of each feature; Determine the path deviation from the real-time risk value across stages, including:

[0063] in, For the stage The path deviates from the real-time risk value. as well as To preset weights, The time decay coefficient, For the first Each stage of diagnosis and treatment and the first Risk transmission matrix for each stage of diagnosis and treatment.

[0064] The optimal classification threshold for the model can be determined based on the ROC curve, enabling a three-level hierarchical early warning system. Low risk: Routine monitoring, no early warning notification; Medium risk: A yellow alert is issued, indicating the top three core causal risk factors and providing suggestions for proactive intervention. High risk: A red alert is issued, and the information is immediately sent to the attending medical staff, simultaneously initiating the deviation tracing and intervention process.

[0065] Step S14: Construct a full-link causal source map based on the structural causal model, and perform hierarchical location and tracing of deviations from the main cause.

[0066] Specifically, it includes: Construct a structural causal model, which includes decision nodes, action nodes, event nodes, and result nodes.

[0067] D-node (decision node): Clinical diagnosis and treatment decisions, including medication selection, ordering of examinations, development of surgical plans, and adjustment of nursing plans; Node A (Action Node): The actual action to be performed corresponding to the decision, including drug administration, nursing procedures, examination execution, path node operations, etc. E-node (event node): The result event corresponding to the action, including changes in the patient's physiological indicators, adverse drug reactions, complications, completion status of pathway nodes, compliance results of diagnosis and treatment behavior, etc. Node O (Outcome Node): Clinical pathway deviation event. For deviation to occur, To ensure compliance with the path.

[0068] Based on the PC causal discovery algorithm combined with expert rules, a directed acyclic graph (DAG) is constructed between nodes. The weight of the causal edge is the CATE causal effect value, which ensures that the association of the edge is a causal relationship, thus solving the source noise problem caused by the association analysis of existing technologies.

[0069] Based on the structural causal model, a full-link causal source map is constructed. ,in, For a full-link causal traceability map, For a set of nodes, It is a set of causal edges.

[0070] When clinical pathway deviation events are detected ( When this occurs, the tracing process is automatically initiated, tracing back from the constructed structural causal model to... Node to root Construct a full-link causal origin graph by identifying all valid causal paths of the nodes. ,in For the set of nodes (D / A / E / O). It is a set of causal edges.

[0071] A depth-first search (DFS) algorithm is used to search for causal paths, retaining only causal effect values. By identifying the causal link and eliminating irrelevant concomitant factors, the clinical effectiveness of the tracing pathway can be ensured.

[0072] Based on the constructed causal origination graph, the total causal effect of each root decision node on the deviation event is calculated, thereby achieving accurate hierarchical positioning of the causes of deviation.

[0073] Based on the end-to-end causal attribution map, the causal effects corresponding to each node are determined, including:

[0074] in, For nodes The causal effect value, For the node To the deviation result All valid causal pathways, The weight of each causal edge in the path. For an effective causal path, It is a causal side.

[0075] according to The values ​​are sorted from largest to smallest, and the reasons for deviation are divided into three levels: The top-ranked node is the root cause of the path deviation; The nodes ranked Top 2-Top 3 are secondary factors that exacerbate path deviation; The nodes are merely accompanying events of the deviation event, not the cause of the deviation.

[0076] Simultaneously, the complete causal path, timestamps of each node, decision-makers, execution records, and clinical evidence are all retained to generate an immutable compliance traceability log, which fully records the entire logical chain of the evolution of diagnosis and treatment decisions, solving the core problems of lack of continuity in decision-making and insufficient medical compliance in existing technologies.

[0077] Step S15: Based on the two-level clinical pathway deviation risk warning model, the full-link causal traceability map, and the results of hierarchical positioning and tracing, a personalized correction plan generation model is constructed. The personalized correction plan generation model is used to generate clinical correction plans.

[0078] Specifically, it includes: Based on the causal effect values ​​of each node, the path deviation is classified into levels, including fully reversible, partially reversible, and irreversible.

[0079] Fully reversible: Deviations that do not cause adverse clinical effects on patients can be completely reversed to standard clinical pathways through simple interventions; Partially reversible: Deviations have caused minor adverse effects / changes in the condition, requiring standardized intervention and control, and can be partially reversed back to the standard clinical pathway; Irreversible: Deviation has caused serious adverse events / significant deterioration of the condition, making it impossible to return to the original standard pathway, and a new personalized clinical pathway needs to be developed.

[0080] A deep reinforcement learning model based on a deep Q-network is constructed, and the deep reinforcement learning model is used as a model for generating personalized correction schemes. The deep reinforcement learning model includes an agent, environment, state space, and action space.

[0081] Agent: Clinical pathway deviation management decision system.

[0082] Environment: Patient's clinical status, hospital medical resource allocation, clinical pathway standardization, and medical insurance DRG payment policy constraints.

[0083] State Space: Patient baseline feature vector, deviation reversibility classification, root cause analysis, current stage of diagnosis and treatment, and availability of medical resources.

[0084] Action Space: All corrective interventions that comply with clinical guidelines, including supplementary examinations, medication regimen adjustments, enhanced nursing monitoring, multidisciplinary consultations, and pathway adjustments, are reviewed by clinical experts to ensure compliance.

[0085] The reward function for constructing the personalized correction scheme generation model includes:

[0086] in, As a reward value, All are weights. For path compliance rewards, Rewards for patient safety Incentives for improving treatment efficiency As a reward for medical costs; Personalized correction schemes are generated based on a personalized correction scheme generation model.

[0087] After the model generates an initial correction plan, it is further optimized by taking into account the patient's individual circumstances (age, comorbidities, economic status, and personal treatment preferences) and the hospital's medical resource allocation to ensure the feasibility and suitability of the plan.

[0088] The correction plan is visualized through the AI ​​central platform, including the correction goal, intervention actions, execution time, responsible person, monitoring indicators, and evidence-based medicine basis.

[0089] During the implementation of the program, the intervention effect is evaluated every 24 hours. If the expected goal is not achieved, the program is optimized again. After the implementation is completed, the clinical pathway is updated, and the entire decision-making process, intervention actions, and effect evaluation are recorded in the traceability log to form an execution closed loop.

[0090] It also includes: a personalized correction scheme generation model based on 10-fold cross-validation optimization.

[0091] Specifically, after each path deviation event is handled, the system automatically collects feedback data from the entire process, including: the model's risk prediction accuracy, the accuracy of tracing the root cause, the clinical effectiveness of the correction plan, the correction data manually annotated by clinicians, the optimal correction plan confirmed by clinical experts, and the patient's clinical outcome data. Monthly incremental training of the model is performed, adding new feedback data to the training set, optimizing the tree structure of the temporal causal forest, the edge weights of the causal origination graph, the reward function and action space of the deep reinforcement learning model, and optimizing the model hyperparameters through ten-fold cross-validation to ensure continuous improvement in model performance. Every quarter, based on the latest clinical practice guidelines, medical insurance policies, hospital management standards, and clinical research evidence, we update the stage feature database, treatment stage division rules, risk warning thresholds, causal rules, and corrective action knowledge base to ensure that the model always meets the latest clinical and management requirements.

[0092] In summary, this invention, by integrating multi-source heterogeneous clinical data and performing standardized processing, achieves unified modeling and efficient utilization of information throughout the entire diagnosis and treatment process, significantly improving the completeness and consistency of data utilization. Through temporal stage segmentation and the construction of stage-specific risk feature databases, key risk factors at different stages of diagnosis and treatment are accurately characterized, thereby enhancing the model's ability to identify pathway deviation risks. Furthermore, the introduction of a temporal causal forest model combined with a cross-stage risk prediction mechanism enables real-time assessment and prospective prediction of clinical pathway deviation risks, improving the timeliness and accuracy of risk warnings.

[0093] This invention constructs a full-link causal source map based on a structural causal model, enabling multi-level localization and tracing of key causes of path deviations, thus enhancing the interpretability of the analysis results. By combining two-level risk warning results with causal source information, a personalized correction plan generation model is constructed, providing targeted intervention strategies for different patients. This effectively reduces the incidence of path deviations, improves medical quality and safety, and optimizes the efficiency of medical resource allocation.

[0094] Based on the same inventive concept, the present invention provides, as follows: Figure 2 The clinical pathway deviation risk management and traceability device shown includes: The acquisition module 21 is used to acquire multi-source heterogeneous clinical data and preprocess the multi-source heterogeneous clinical data, including time period alignment and data standardization. The risk feature library construction module 22 is used to divide time-series stages based on preprocessed multi-source heterogeneous clinical data and construct a stage-specific risk feature library. The risk warning model construction module 23 is used to construct a time-series causal forest model based on a stage-specific risk feature library, and to construct a two-level clinical pathway deviation risk warning model by combining a real-time risk cross-stage prediction mechanism. The graph construction module 24 is used to construct a full-link causal source map based on the structural causal model, and to perform hierarchical location and tracing of deviations from the main cause; The scheme generation module 25 is used to construct a personalized correction scheme generation model based on the two-level clinical pathway deviation risk warning model, the full-link causal traceability map, and the results of hierarchical positioning and tracing. The personalized correction scheme generation model is used to generate clinical correction schemes.

[0095] Based on the same inventive concept, the present invention also provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to execute a clinical pathway deviation risk management and tracing method based on big data and AI, as described above.

[0096] Based on the same inventive concept, the present invention also provides a non-transitory computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor of an electronic device, enables the electronic device to execute the aforementioned clinical pathway deviation risk management and tracing method based on big data and AI.

[0097] Since the electronic device described in this embodiment is an electronic device used to implement the information processing method in the embodiments of the present invention, those skilled in the art can understand the specific implementation methods and various variations of the electronic device in this embodiment based on the information processing method described in the embodiments of the present invention. Therefore, how the electronic device implements the method in the embodiments of the present invention will not be described in detail here. Any electronic device used by those skilled in the art to implement the information processing method in the embodiments of the present invention falls within the scope of protection of the present invention.

[0098] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0099] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0100] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0101] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the function specified in one or more boxes.

[0102] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0103] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A clinical pathway deviation risk management and tracing method based on big data and AI, characterized in that, include: Acquire multi-source heterogeneous clinical data and preprocess the multi-source heterogeneous clinical data, wherein the preprocessing includes time period alignment and data standardization; Based on the preprocessed multi-source heterogeneous clinical data, time-series stages were divided, and a stage-specific risk feature library was constructed. Based on the stage-specific risk feature library, a time-series causal forest model is constructed, and combined with a real-time risk cross-stage prediction mechanism, a two-level clinical pathway deviation risk early warning model is built, including: Based on preprocessed multi-source heterogeneous clinical data, a causal inference standard sample set is constructed, wherein the causal inference standard sample set includes sample units, label definitions, and causal processing variables; Based on the aforementioned causal inference standard sample set and the stage-specific risk feature library, a time-series causal forest model is constructed, including: in, For the stage The Middle The causal effect value of each feature, As expected, The path deviation label represents the result after the deviation has occurred. The result of labeling path deviation as path compliance For the stage Features; Based on the causal splitting criterion, a splitting gain is applied to the aforementioned temporal causal forest model, including: in, For split gain, The CATE value of the parent node. The CATE value of the left child node after the split. This is the CATE value of the right child node after the split; Determine the path deviation from the real-time risk value at the current stage of diagnosis and treatment, including: in, For the stage The path deviates from the real-time risk value. For the stage The Middle The weights of each feature For the stage The number of features; Determine the path deviation from the real-time risk value across stages, including: in, For the stage The path deviates from the real-time risk value. as well as To preset weights, The time decay coefficient, For the first Each stage of diagnosis and treatment and the first Risk transmission matrix for each stage of diagnosis and treatment; A full-link causal source map is constructed based on a structural causal model, and the deviation from the main cause is hierarchically located and traced. The structural causal model includes decision nodes, action nodes, event nodes and result nodes. Based on the aforementioned two-level clinical pathway deviation risk warning model, the aforementioned full-link causal tracing map, and the results of hierarchical positioning and tracing, a personalized correction plan generation model is constructed, wherein the personalized correction plan generation model is used to generate clinical correction plans.

2. The clinical pathway deviation risk management and tracing method based on big data and AI as described in claim 1, characterized in that, Based on preprocessed multi-source heterogeneous clinical data, time-series phases were segmented, and a phase-specific risk feature library was constructed, including: Dividing the diagnosis and treatment stages ,in, This is a collection of diagnostic and treatment stages. For the first Each stage of diagnosis and treatment; Construct feature sets corresponding to each stage of diagnosis and treatment, including standard pathway compliance features, patient physiological state features, and clinical event risk features; Determine the weight of each feature, including: in, For the stage The Middle The weights of each feature Information gain ratio For the stage The Middle One characteristic, Labels for path deviation; Construct a risk transmission matrix across treatment stages, where the dimensions of the risk transmission matrix are: Dimensions include: in, For the first Each stage of diagnosis and treatment and the first Risk transmission matrix for each stage of diagnosis and treatment For conditional probability, For the first The path deviates from the label at each stage of diagnosis and treatment. For the first The path deviates from the label at each stage of diagnosis and treatment.

3. The clinical pathway deviation risk management and tracing method based on big data and AI as described in claim 1, characterized in that, Based on a structural causal model, a full-link causal source map is constructed, and deviations from the main cause are hierarchically located and traced, including: Construct a structural causal model, which includes decision nodes, action nodes, event nodes, and result nodes; Based on the aforementioned structural causal model, a full-link causal tracing graph is constructed. ,in, For a full-link causal traceability map, For a set of nodes, It is a set of causal edges; Based on the end-to-end causal attribution map, the causal effects corresponding to each node are determined, including: in, For nodes The causal effect value, For the node To the deviation result All valid causal pathways, The weight of each causal edge in the path. For an effective causal path, It is a causal side.

4. The clinical pathway deviation risk management and tracing method based on big data and AI as described in claim 1, characterized in that, Based on the aforementioned two-tiered clinical pathway deviation risk warning model, the aforementioned full-link causal tracing map, and the results of hierarchical localization and tracing, a personalized correction plan generation model is constructed, including: Based on the causal effect values ​​of each node, the path deviation is classified into levels, including fully reversible, partially reversible, and irreversible. A deep reinforcement learning model based on a deep Q-network is constructed, and the deep reinforcement learning model is used as a model for generating personalized correction schemes. The deep reinforcement learning model includes an agent, environment, state space, and action space. The reward function for constructing the personalized correction scheme generation model includes: in, As a reward value, All are weights. For compliance rewards, As a reward for patient safety, Incentives for improving treatment efficiency As a reward for medical costs; A personalized correction scheme is generated based on the personalized correction scheme generation model.

5. The clinical pathway deviation risk management and tracing method based on big data and AI as described in claim 1, characterized in that, Preprocessing of the aforementioned multi-source heterogeneous clinical data includes: Timeline alignment of high-frequency continuous time-series data, discrete clinical event data, and static baseline data; Based on Z-score standardization, data standardization is performed on multi-source heterogeneous clinical data.

6. The clinical pathway deviation risk management and tracing method based on big data and AI as described in claim 1, characterized in that, Also includes: The personalized correction scheme generation model is optimized based on 10-fold cross-validation.

7. A clinical pathway deviation risk management and traceability device based on big data and AI, characterized in that, The clinical pathway deviation risk management and tracing method based on big data and AI, applied to any one of claims 1-6, includes: The acquisition module is used to acquire multi-source heterogeneous clinical data and preprocess the multi-source heterogeneous clinical data, wherein the preprocessing includes time period alignment and data standardization. The risk feature library construction module is used to divide time-series phases based on preprocessed multi-source heterogeneous clinical data and construct phase-specific risk feature libraries. The risk warning model construction module is used to construct a time-series causal forest model based on the stage-specific risk feature library, and to construct a two-level clinical pathway deviation risk warning model by combining a real-time risk cross-stage prediction mechanism. The graph construction module is used to construct a full-link causal source map based on the structural causal model, and to perform hierarchical location and tracing of deviations from the main cause; The scheme generation module is used to construct a personalized correction scheme generation model based on the two-level clinical pathway deviation risk warning model, the full-link causal tracing map, and the results of hierarchical positioning and tracing. The personalized correction scheme generation model is used to generate clinical correction schemes.

8. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the clinical pathway deviation risk management and tracing method based on big data and AI as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is able to execute the clinical pathway deviation risk management and tracing method based on big data and AI as described in any one of claims 1 to 6.

Images