Monitoring methods, devices, and storage media for occult symptoms during the recovery period of acute illnesses in the elderly.

Abstract

This application discloses a method, device, and storage medium for monitoring hidden conditions during the recovery period of acute illnesses in the elderly, relating to the field of medical information processing technology. This application improves the quality of multi-source health data through dual-driven data collection based on the recovery stage and relapse risk, dual-dimensional data conflict resolution, and cross-subtype data enhancement. By combining spatial and temporal features, a dual-output model, and a temporal causal algorithm, it achieves accurate prediction of relapse risk and intervention effectiveness, constructing a closed-loop hierarchical intervention program. This effectively identifies hidden conditions during the recovery period of acute illnesses in the elderly, improves the accuracy of early warning and the effectiveness of intervention, and reduces the risk of relapse.

Description

Technical Field

[0001] This application relates to the field of medical information processing technology, specifically to a method, device, and storage medium for monitoring hidden symptoms during the recovery period of acute illness in the elderly. Background Technology

[0002] With the increasing trend of population aging, health management of the elderly has become a core challenge in the field of public health. The elderly often have multiple underlying diseases, and during the recovery period after treatment for acute illnesses (such as cardiovascular emergencies, infectious emergencies, and metabolic disorder emergencies), their physiological compensatory capacity is weak, their condition fluctuates insidiously, and early signs are atypical, making them highly susceptible to relapses, deterioration, or even sudden critical events during the recovery phase. Traditional monitoring during the recovery period relies heavily on regular in-hospital checkups, simple home monitoring, or manual recording, which has significant limitations.

[0003] On the one hand, traditional health monitoring often adopts a fixed frequency, fixed indicators, and fixed equipment collection mode, without dynamic scheduling based on the patient's actual recovery stage and relapse risk level. This easily leads to insufficient monitoring of high-risk patients and excessive collection of data from low-risk patients. At the same time, the triggering mechanism for abnormal indicators is relatively simple, often using a single fixed threshold for instantaneous judgment. It lacks a graded and linked collection strategy that matches the type of acute illness and individual baseline, making it difficult to detect the hidden precursors of multi-indicator synergistic changes in the early stages.

[0004] Furthermore, elderly patients recovering from acute illnesses exhibit diverse acute illness types, complex underlying medical histories, and significant differences in rehabilitation stages. The sample size for some subtypes is sparse, making direct modeling prone to overfitting and poor generalization. Simultaneously, health indicators exhibit distinct temporal characteristics, with varying temporal contributions from different rehabilitation stages and different indicator types. Traditional data integration methods often employ static weights or simple time decay, failing to fully uncover early risk signals within temporal changes.

[0005] At the level of disease assessment and intervention, existing technologies primarily rely on single-dimensional risk level prediction, failing to simultaneously output relapse risk and intervention effect predictions, leading to a disconnect between risk warnings and clinical intervention. Furthermore, they often remain at the level of indicator correlation analysis, struggling to uncover the temporal causal transmission relationships between multiple abnormal indicators, and unable to pinpoint the true starting point and key targets driving disease deterioration. Intervention measures are often empirical and generalized, lacking specificity and effectiveness. Traditional intervention programs are mostly fixed or simple tiered programs, failing to meet the actual clinical needs of elderly patients during the recovery period, who exhibit variable disease conditions, significant individual differences, and inconsistent compliance. This hinders the early identification, continuous monitoring, and improvement of rehabilitation quality for hidden conditions. Summary of the Invention

[0006] The purpose of this application is to provide a method, device, and storage medium for monitoring hidden conditions during the recovery period of acute illnesses in the elderly, in order to solve the problem of low accuracy in traditional monitoring of hidden conditions during the recovery period of acute illnesses in the elderly.

[0007] To achieve the above objectives, the first aspect of this application provides a method for monitoring latent symptoms during the recovery period of acute illness in the elderly, comprising: Based on the dual-drive data collection and scheduling mechanism of elderly patients’ rehabilitation stage and relapse risk level, health indicator monitoring data is obtained, and when health indicators are detected to exceed the baseline range, the hierarchical linkage collection of specific health indicators is triggered to obtain multi-source health data. Based on the medical knowledge base benchmark and individual dynamic baseline, the multi-source health data is subject to two-dimensional data conflict adjudication. Patient subtypes are classified according to acute type, underlying medical history and rehabilitation stage. Cross-subtype transfer enhancement is performed on the multi-source health data. The time decay weight is dynamically adjusted in combination with rehabilitation stage and indicator type to obtain integrated data. The spatial correlation features and temporal dependency features of the integrated data are extracted and input into a preset dual-output model to obtain the relapse risk prediction value and the intervention effect prediction value. The temporal causal relationship chain and intervention target of the compound precursor are mined by the temporal causal discovery algorithm, and the temporal causal relationship chain and intervention target are optimized based on the relapse risk prediction value. Based on the optimized temporal causal relationship chain and intervention targets, a tiered intervention plan is generated. The tiered intervention plan is dynamically adjusted by combining the intervention implementation rate, the health indicator achievement rate and the predicted value of the intervention effect, and the evaluation results of the tiered intervention plan are obtained.

[0008] The second aspect of this application provides a monitoring device for insidious symptoms during the recovery period of an elderly patient with an acute illness, comprising: The triggering module is used to acquire health indicator monitoring data based on the dual-drive collection and scheduling mechanism of the elderly patient's rehabilitation stage and relapse risk level. When the health indicator is detected to be outside the baseline range, it triggers the hierarchical linkage collection of specific health indicators to obtain multi-source health data. The integration module is used to perform two-dimensional data conflict adjudication on the multi-source health data based on the medical knowledge base benchmark and individual dynamic baseline, and to classify patients into subtypes according to acute type, basic medical history and rehabilitation stage, and to perform cross-subtype transfer enhancement on the multi-source health data. The time decay weight is dynamically adjusted in combination with rehabilitation stage and indicator type to obtain integrated data. An optimization module, configured to extract the spatial association features and temporal dependence features of the integrated data and input them into a preset dual-output model to obtain a recurrence risk prediction value and an intervention effect prediction value, use a temporal causal discovery algorithm to mine the temporal causal relationship chain and intervention targets of composite precursors, and optimize the temporal causal relationship chain and intervention targets based on the recurrence risk prediction value; An evaluation module, configured to generate a hierarchical intervention plan based on the optimized temporal causal relationship chain and intervention targets, dynamically adjust the hierarchical intervention plan in combination with the intervention execution degree, the health index compliance rate, and the intervention effect prediction value, and obtain the evaluation result of the hierarchical intervention plan.

[0009] In a third aspect of the present application, a computer-readable storage medium is provided. A program is stored in the computer-readable storage medium, and the program can be loaded and executed by a processor to perform the above-mentioned method for monitoring occult conditions during the rehabilitation period of elderly acute diseases.

[0010] The beneficial effects of the present application are as follows: Through the dual-drive acquisition scheduling mechanism of the rehabilitation stage and the recurrence risk level, and triggering hierarchical linkage acquisition when indicators are abnormal, the present application can accurately obtain multi-source health data, improve the monitoring sensitivity and data acquisition efficiency of occult conditions during the rehabilitation period of elderly acute diseases. By using the dual-dimensional data conflict adjudication of the medical knowledge base benchmark and the individual dynamic baseline, combined with cross-subtype migration enhancement and dynamic time decay weights, it can effectively eliminate data noise and individual differences, and improve the integration quality and reliability of multi-source health data. By extracting spatial association features and temporal dependence features, and using a dual-output model to realize recurrence risk and intervention effect prediction, combined with a temporal causal discovery algorithm to mine the temporal causal relationship chain and intervention targets, it can identify occult condition changes earlier, improve the warning accuracy and prediction ability. Generate a hierarchical intervention plan based on the optimized temporal causal relationship chain and intervention targets, and dynamically adjust according to the intervention execution degree, the index compliance rate, and the intervention effect prediction value to form a closed-loop optimization mechanism, improve the pertinence and effectiveness of the intervention plan, reduce the recurrence risk, and improve the rehabilitation effect of elderly patients.

[0011] Other features and advantages of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a schematic diagram of an application scenario of a method for monitoring occult conditions during the rehabilitation period of elderly acute diseases provided in an embodiment of the present application; Figure 2 It is a schematic flowchart of a method for monitoring occult conditions during the rehabilitation period of elderly acute diseases provided in an embodiment of the present application; Figure 3 It is a schematic structural diagram of a device for monitoring occult conditions during the rehabilitation period of elderly acute diseases provided in an embodiment of the present application. Detailed Implementation

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified. Details are set forth in the following description for illustrative purposes. It should be understood that those skilled in the art will recognize that this application can be implemented without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid unnecessarily obscuring the description of this application. Therefore, this application is not intended to be limited to the embodiments shown, but rather to be consistent with the broadest scope of the principles and features disclosed herein.

[0015] The monitoring method for hidden conditions during the recovery period of acute illness in the elderly in this application embodiment is applied to a monitoring device for hidden conditions during the recovery period of acute illness in the elderly, and the monitoring device for hidden conditions during the recovery period of acute illness in the elderly is set in an electronic device. For example Figure 1 As shown, Figure 1 This is a schematic diagram illustrating an application scenario of the monitoring method for hidden conditions during the recovery period of acute illness in the elderly, as described in this application. The application scenario of the monitoring method for hidden conditions during the recovery period of acute illness in the elderly includes an electronic device 110 for monitoring hidden conditions during the recovery period of acute illness in the elderly. The electronic device 110 integrates a monitoring device for hidden conditions during the recovery period of acute illness in the elderly to run a computer-readable storage medium corresponding to the monitoring method for hidden conditions during the recovery period of acute illness in the elderly, so as to execute the steps of the monitoring method for hidden conditions during the recovery period of acute illness in the elderly.

[0016] Understandable, Figure 1The electronic devices in the application scenario of the monitoring method for monitoring the concealed condition during the recovery period of acute illness in the elderly, or the devices contained in the electronic devices, do not constitute a limitation on the embodiments of this application. That is, the number or type of devices in the application scenario of the monitoring method for monitoring the concealed condition during the recovery period of acute illness in the elderly, or the number or type of devices contained in each device, do not affect the overall implementation of the technical solution in the embodiments of this application, and can all be considered as equivalent substitutions or derivatives of the technical solutions claimed in the embodiments of this application.

[0017] In this application embodiment, the electronic device 110 can be an independent device, or a device network or device cluster composed of devices. For example, the electronic device 110 described in this application embodiment includes, but is not limited to, a computer, a network host, a single network device, a set of multiple network devices, or a cloud device composed of multiple devices. Among them, the cloud device is composed of a large number of computers or network devices based on cloud computing.

[0018] Those skilled in the art will understand that Figure 1 The application scenarios shown are merely one application scenario corresponding to the technical solution of this application, and do not constitute a limitation on the application scenarios of the technical solution of this application. Other application scenarios may include more than one application scenario. Figure 1 The number of more or fewer electronic devices shown, or the network connections of electronic devices, for example Figure 1 Only one electronic device is shown in the image. It is understood that the scenario of this monitoring method for concealed symptoms during the recovery period of acute illness in the elderly may also include one or more other electronic devices, which are not specifically limited here. The electronic device 110 may also include a memory and a processor. The memory is used to store information related to the monitoring method for concealed symptoms during the recovery period of acute illness in the elderly.

[0019] Furthermore, in the application scenario of the monitoring method for concealed symptoms during the recovery period of acute illness in the elderly according to this application embodiment, the electronic device 110 can be equipped with a display device, or the electronic device 110 can be without a display device but connected to an external display device 120. The display device 120 is used to output the results of the monitoring method for concealed symptoms during the recovery period of acute illness in the elderly executed by the electronic device. The electronic device 110 can access the background database 130. The background database 130 can be the local storage of the electronic device 110 or a cloud database located in the cloud. The background database 130 stores information related to the monitoring method for concealed symptoms during the recovery period of acute illness in the elderly.

[0020] It should be noted that, Figure 1 The application scenario of the monitoring method for hidden symptoms during the recovery period of acute illness in the elderly shown is merely an example. The application scenario of the monitoring method for hidden symptoms during the recovery period of acute illness in the elderly described in this application embodiment is to more clearly illustrate the technical solution of this application embodiment and does not constitute a limitation on the technical solution provided in this application embodiment.

[0021] Based on the application scenarios of the aforementioned monitoring methods for hidden symptoms during the recovery period of acute illnesses in the elderly, an embodiment of the monitoring method for hidden symptoms during the recovery period of acute illnesses in the elderly is proposed. A detailed description is provided below with reference to the accompanying drawings.

[0022] Figure 2 This is a flowchart illustrating a method for monitoring latent symptoms in elderly patients during the recovery period of acute illness, as provided in this application embodiment. Figure 2 As shown, this monitoring method can be executed by the processor in the above-mentioned electronic device 110, and steps 201-204 are described in detail below.

[0023] Step 201: Based on the dual-drive data collection and scheduling mechanism of the elderly patient's rehabilitation stage and relapse risk level, acquire health indicator monitoring data, and trigger the hierarchical linkage collection of specific health indicators when the health indicators are detected to exceed the baseline range, so as to obtain multi-source health data.

[0024] The rehabilitation phase refers to the period of recovery defined based on the treatment progress and physiological recovery of elderly patients with acute illnesses. It includes the acute phase (1-4 weeks), the recovery phase (5-12 weeks), and the stable phase (13 weeks+). The relapse risk level is a pre-defined risk level based on factors such as the patient's age, underlying diseases, type of acute illness, and history of relapses. It includes low risk (S<0.3), medium risk (0.3≤S<0.7), and high risk (S≥0.7).

[0025] The dual-drive data collection and scheduling mechanism refers to dynamically determining the collection frequency, type, and duration of health indicators based on the recovery stage and relapse risk level, achieving on-demand data collection. When routine health indicators exceed the preset benchmark range, tiered linkage data collection is automatically triggered. That is, according to the severity of abnormal indicators, corresponding specialized indicators, multimodal devices, and multi-time period data are collected collaboratively at each level, thereby obtaining multi-source health data covering physiological indicators, activity status, medication status, and other dimensions and sources.

[0026] By employing the above methods, we can ensure the continuity of monitoring while reducing data redundancy and equipment burden caused by excessive data collection, and improve the ability to capture hidden abnormal signals.

[0027] Step 202: Based on the medical knowledge base benchmark and individual dynamic baseline, perform two-dimensional data conflict adjudication on multi-source health data, and classify patients into subtypes according to acute type, underlying medical history and rehabilitation stage. Perform cross-subtype transfer enhancement on multi-source health data, and dynamically adjust the time decay weight in combination with rehabilitation stage and indicator type to obtain integrated data.

[0028] The medical knowledge base benchmark is a standard reference system composed of industry-standard medical guidelines, normal values ​​for laboratory tests, and emergency diagnosis and treatment guidelines.

[0029] Dual-dimensional data conflict resolution refers to the simultaneous use of general medical standards and patient-specific baselines to verify, denoise, complete, and determine conflicts in multi-source health data, thereby eliminating abnormal noise data and improving data reliability.

[0030] Patient subtyping involves a refined classification of patients based on dimensions such as acute illness type, underlying medical history, and recovery stage. Cross-subtype transfer enhancement utilizes the data distribution characteristics of patients in similar subtypes to enhance and supplement data with insufficient sample size or sparse features within the current subtype, improving the data's generalization ability. Time decay weighting assigns dynamic weights to data at different time points based on changes in recovery stage and the importance of indicators; recent key data has higher weights, while the weight of longer-term data gradually decreases, highlighting the trend of time-series changes.

[0031] The above processing can effectively solve the problems of heterogeneous, sparse, noisy, and significant individual differences in elderly patient data, and obtain high-quality and highly usable integrated data.

[0032] Step 203: Extract the spatial correlation features and temporal dependency features of the integrated data and input them into the preset dual-output model to obtain the predicted value of relapse risk and the predicted value of intervention effect. Use the temporal causal discovery algorithm to mine the temporal causal relationship chain and intervention target of the compound precursor, and optimize the temporal causal relationship chain and intervention target based on the predicted value of relapse risk.

[0033] Spatial correlation characteristics reflect the mutual correlation and coupling relationship between multiple health indicators at the same time; temporal dependence characteristics reflect the evolution pattern and lag relationship of the same indicator or multiple indicators in the time series.

[0034] The dual-output model is a pre-trained deep learning model that can simultaneously output relapse risk prediction and intervention effect prediction, enabling parallel risk prediction and effect evaluation.

[0035] The temporal causal discovery algorithm is used to mine the causal relationships between multiple hidden precursors from temporal health data, form a temporal causal relationship chain, and locate key nodes that can be changed through intervention, i.e., intervention targets.

[0036] Combining relapse risk prediction values ​​with the screening and optimization of temporal causal chains and intervention targets can improve the accuracy of early warning and the targeting of interventions, achieving an upgrade from correlation analysis to causal judgment.

[0037] Step 204: Based on the optimized temporal causal relationship chain and intervention targets, generate a stratified intervention plan, and dynamically adjust the stratified intervention plan by combining the intervention implementation rate, the health indicator achievement rate and the predicted value of the intervention effect, and obtain the evaluation results of the stratified intervention plan.

[0038] The stratified intervention plan is a hierarchical, classified, and time - segmented intervention strategy formulated according to the recurrence risk level, rehabilitation stage, and patient subtype, including content such as adjustment of monitoring frequency, medication reminder, lifestyle guidance, and emergency warning.

[0039] The intervention implementation degree is used to measure the actual implementation of the intervention measures; the health index compliance rate is used to reflect the improvement degree of relevant indicators after the intervention.

[0040] In this step, the intervention implementation degree, the index compliance rate, and the predicted value of the intervention effect are combined to perform real - time iteration and dynamic adjustment on the stratified intervention plan, forming a closed - loop system of "monitoring - recognition - prediction - intervention - evaluation - optimization".

[0041] The finally output evaluation results can be used to guide the adjustment of subsequent rehabilitation strategies, improve the management level of hidden conditions during the rehabilitation period of elderly acute diseases, reduce the recurrence probability, and enhance the rehabilitation quality and safety.

[0042] In the embodiment of this application, through the dual - drive acquisition and scheduling mechanism driven by the rehabilitation stage and the recurrence risk level, and triggering hierarchical linkage acquisition when the indicators are abnormal, it can accurately obtain multi - source health data, improve the monitoring sensitivity and data acquisition efficiency of hidden conditions during the rehabilitation period of elderly acute diseases. By using the dual - dimension data conflict adjudication of the medical knowledge base benchmark and the individual dynamic baseline, combined with cross - subtype migration enhancement and dynamic time - decay weight, it can effectively eliminate data noise and individual differences, and improve the integration quality and reliability of multi - source health data. By extracting spatial association features and temporal dependence features, and using a dual - output model to predict the recurrence risk and intervention effect, combined with the temporal causal discovery algorithm to mine the temporal causal relationship chain and intervention targets, it can identify the changes of hidden conditions earlier, improve the warning accuracy and prediction ability. Based on the optimized temporal causal relationship chain and intervention targets, a stratified intervention plan is generated, and it is dynamically adjusted according to the intervention implementation degree, the index compliance rate, and the predicted value of the intervention effect, forming a closed - loop optimization mechanism, enhancing the pertinence and effectiveness of the intervention plan, reducing the recurrence risk, and improving the rehabilitation effect of elderly patients.

[0043] In step 201, for the step of obtaining health index monitoring data through the dual - drive acquisition and scheduling mechanism, first, the rehabilitation stage of elderly patients can be divided into acute - phase rehabilitation, recovery period, and stable period according to clinical standards. The rehabilitation stage is a period divided according to the degree of physiological function recovery and the degree of disease stability after the treatment of elderly acute - disease patients. According to the general clinical acute - disease rehabilitation diagnosis and treatment standards, the rehabilitation period of elderly patients is divided into stages, and the corresponding physiological states, monitoring focuses, and intervention intensities in different stages are clarified, providing a unified stage basis for subsequent differential monitoring and intervention, and enhancing the clinical adaptability of the monitoring plan.

[0044] Based on real-time health indicator monitoring data of elderly patients, a pre-built initial risk model is used for analysis and calculation to output the relapse risk level. The initial risk model is pre-trained and constructed based on a large number of elderly acute illness rehabilitation cases, health indicators, and relapse events, and is used to quickly output the relapse risk level assessment model. The relapse risk level is a grade label used to characterize the likelihood of relapse in elderly patients in the acute illness rehabilitation period, and can include low, medium, and high relapse risk levels. Through automated and quantitative assessment of relapse risk, the problems of strong subjectivity and low efficiency in manual assessment are reduced, providing an objective basis for subsequent dynamic scheduling.

[0045] Next, based on the combination relationship between the recovery stage and the relapse risk level, the collection frequency, indicator priority, and equipment coordination logic of health indicators are dynamically adjusted. By combining and matching the recovery stage with the relapse risk level, the collection frequency, indicator monitoring priority, and the working mode of the monitoring equipment are adaptively adjusted according to different combination scenarios to achieve a precise match between the monitoring strategy and the patient's actual condition, ensuring sufficient monitoring for high-risk patients while reducing resource waste for low-risk patients.

[0046] The frequency of health indicator collection gradually decreases as the recovery phase progresses and gradually increases as the relapse risk level rises. In other words, as patients move from the acute phase to the recovery and stabilization phases, the collection frequency gradually decreases; conversely, as the relapse risk level increases, the collection frequency increases accordingly. The collection frequency is the number of times health indicators are collected per unit of time, reflecting the monitoring density. Furthermore, at high relapse risk levels, all monitoring devices are activated and work collaboratively, while at low risk levels, unnecessary monitoring devices are shut down, significantly reducing system power consumption and data redundancy. Simultaneously, it ensures comprehensive monitoring in high-risk scenarios, balancing energy efficiency and safety.

[0047] See Table 1, which is an example of a dual-drive acquisition scheduling mechanism.

[0048] Table 1

[0049] The dynamic calculation model for indicator priority is used to calculate the first The real-time priority of each indicator determines the weight of the data collection resource allocation. As an example, the dynamic calculation model for indicator priority can satisfy: ; in, The medical importance coefficient (based on the "Clinical Guidelines for Rehabilitation of Geriatric Emergency Diseases", with a value of 0.1-1.0, and core indicators such as heart rate variability are taken as 1.0). The individual correlation score is the degree of correlation between the indicator and the patient's previous acute illnesses, with a value of 0.1-1.0, such as 1.0 for blood glucose in diabetic patients. The risk sensitivity coefficient (the correlation between abnormal indicators and the risk of recurrence, obtained through training with historical data, with a value of 0.1-1.0) is used. , and The weighting coefficients are optimized based on clinical data. For example, the top 5 priority indicators are allocated 80% of the data collection resources, while subsequent indicators are allocated 20% of the resources, and the calibration is recalculated every 2 weeks.

[0050] In step 201, triggering the hierarchical linkage collection of specific health indicators may include the following steps.

[0051] First, typical abnormal patterns in elderly patients with acute conditions are integrated, and rules for linking basic abnormal signals with specific indicators are formulated, pre-constructing a target trigger rule base. The target trigger rule base is a set of rules storing the correspondence between abnormal signals and specific indicators, as well as the triggering conditions. It includes the mapping relationship between abnormal signals and specific health indicators. Each basic abnormal signal is pre-configured with one or more sets of specific health indicators, forming a one-to-one or one-to-many mapping relationship, enabling rapid matching of abnormal signals to specific indicators, reducing blind data collection, and improving the targeting of specific data collection. Common abnormal indicator change patterns during the rehabilitation period of elderly patients with acute conditions are summarized, and linkage judgment rules between basic abnormal signals and specific indicators are established, forming a directly callable rule base. This provides standardized rule support for automatic anomaly identification and linkage data collection, improving the consistency and response speed of the system's judgment.

[0052] For example, it integrates typical abnormal patterns of 12 types of acute geriatric conditions (myocardial infarction, cerebral infarction, diabetic ketoacidosis, etc.), including 18 rules linking basic signal abnormalities with specific indicators. The individual dynamic baseline is a personalized normal reference range based on the patient's historical health data. Basic signal abnormality determination can include: when an indicator exceeds the individual baseline ±20% (or a medical standard threshold) and lasts for ≥10 minutes, it is considered an abnormal trigger condition. Linkage with specific indicators can include: a sudden increase in heart rate ≥20% → triggering high-frequency collection of CRP / IL-6, blood oxygen, and troponin (5 minutes / time, lasting 2 hours); an increase in sleep apnea duration of 50% → triggering specific collection of blood pressure, PHQ-9 mood score, and cognitive function. Trigger intensity grading can include: abnormality lasting 10-30 minutes as mild trigger (specific indicator collection frequency doubled); and lasting >30 minutes as deep trigger (activating all relevant specific indicators + medical staff alert).

[0053] When a health indicator exceeds an individual's dynamic baseline or a medical knowledge base benchmark, and the duration is greater than or equal to a set duration, it is considered a valid anomaly trigger condition. The valid anomaly trigger condition is the formal criterion used to determine whether to initiate the linked collection of specific indicators. Simultaneously, both the individual baseline and general medical benchmarks are used for judgment, and a duration threshold is added to filter out instantaneous fluctuations; only when the condition is met is it considered a valid anomaly. This reduces false triggers caused by instantaneous interference and misoperation, improves the accuracy of anomaly detection, and reduces the false alarm rate.

[0054] Next, based on the medical importance coefficient of clinical guidelines, the individual correlation between indicators and past acute illnesses, and the risk sensitivity coefficient of indicator abnormalities and recurrence risk, a dynamic indicator priority calculation model was used to determine the collection weights of specific indicators. This dynamic indicator priority calculation model is a calculation model that automatically determines the collection priority of specific indicators based on multi-coefficient weighted calculation. By inputting the coefficients from the three dimensions into the dynamic calculation model, the model outputs the collection weights and priority rankings of each specific indicator, making the collection weights more aligned with clinical value and individual risk characteristics, and prioritizing the collection of indicators with the greatest predictive significance.

[0055] Based on valid abnormal signals, corresponding specific health indicators are matched from the target trigger rule base. Trigger intensity is categorized according to the duration of the abnormality, and alerts are sent to healthcare terminals. Trigger intensity determines the urgency of the coordinated data collection and alert based on the duration of the abnormality. By matching specific indicators from the rule base based on the basic signals, different trigger intensity levels are assigned according to the duration of the abnormality, and tiered alerts are pushed out. This tiered alert system facilitates rapid identification of urgency by healthcare personnel, improving response efficiency.

[0056] Finally, based on the priority of specific indicators, specialized health data from multiple monitoring devices are collected synchronously and then aggregated with basic monitoring data to obtain multi-source health data. Multi-source health data is a multi-dimensional, multi-device health data set composed of basic monitoring data and data collected in conjunction with specialized indicators. Following a priority order, specialized data is collected synchronously from multiple devices, and the basic and specialized data are merged and aggregated to form comprehensive, high-quality multi-source data, providing a reliable data foundation for subsequent data conflict adjudication, risk prediction, and intervention plans.

[0057] In step 202, the core objective is to resolve the issues of multi-source data conflict and sample homogeneity, thereby improving data quality and feature effectiveness. Firstly, multi-source health data can be preprocessed. For example, data cleaning can employ the 3σ principle combined with the Isolation Forest algorithm to remove outliers, and missing data can be filled using a hybrid approach of LSTM time-series prediction and interpolation of similar patients (error ≤ ±5%). Data standardization can be achieved using Z-score normalization to unify the units of measurement. Dynamic feature association can be achieved by constructing a three-dimensional correlation matrix of indicators, time, and health status, and calculating correlation coefficients to capture the dynamic correlations between indicators.

[0058] Next, conflict resolution is performed on the two-dimensional data. Based on a medical knowledge base, baseline reference intervals for each health indicator are determined to construct the medical knowledge base baseline. This baseline is derived from authoritative medical materials such as clinical practice guidelines, laboratory testing standards, and emergency rehabilitation protocols, providing a unified and standardized universal reference standard for health indicators. Simultaneously, based on the historical health data of elderly patients, a sliding window algorithm is used to segment, update, and fit the historical data to determine a real-time dynamic reference area, thus constructing an individual dynamic baseline. This individual dynamic baseline reflects the patient's personalized physiological normal range at different stages of rehabilitation. By combining the medical knowledge base baseline and the individual dynamic baseline to form a two-dimensional judgment basis, the data judgment is ensured to conform to general medical standards while also adapting to the individual physiological differences of elderly patients, improving the accuracy and rationality of the data judgment.

[0059] When a deviation exceeding a set deviation range is detected in the values ​​of the same health indicator obtained from different acquisition devices or data sources, it is determined to be a multi-source data conflict. A two-dimensional data conflict resolution rule is used to handle multi-source data conflicts. First, it verifies whether each set of conflicting data is within the individual's dynamic baseline. If only a single conflicting data point is within the individual's dynamic baseline, it is considered valid data. If all conflicting data points are either within or outside the individual's dynamic baseline, the degree of closeness between each conflicting data point and the medical knowledge base benchmark is calculated, and the value with the highest degree of closeness is considered valid data. If all conflicting data points exceed the range of both the individual's dynamic baseline and the medical knowledge base benchmark, a third-party medical device is activated for data verification. Simultaneously, the conflicting data is marked as high-risk data and an alert is triggered, reminding medical staff to pay attention to potential abnormalities in the patient's condition. This two-dimensional data conflict resolution effectively solves problems such as data inconsistency, noise interference, and abnormal fluctuations caused by multi-source device acquisition, improving the reliability and credibility of health data and reducing the risk of misjudgment and missed diagnoses.

[0060] For example, medical knowledge base benchmarks can integrate authoritative data such as the "Clinical Guidelines for Rehabilitation of Geriatric Emergency Diseases" and the "Clinical Application Standards for Inflammatory Markers" to establish normal reference ranges for indicators. Individual baseline: Based on the patient's health data from the previous 3 months, a dynamic reference interval was constructed using a sliding window algorithm (window size = 7 days). ,formula: , ( Let be the mean value within the window at time t. (Standard deviation). The adjudication process may include: when multiple source data conflict (e.g., blood pressure from device A = 160 / 100, blood pressure from device B = 130 / 85), first determine whether the data is within the individual's baseline: if the data from device B is within the baseline... Within the given timeframe, data from device B will be prioritized; if all data exceed the individual baseline, refer to the medical knowledge base for more relevant data. If the data exceeds the two-dimensional benchmark, a third-party device (such as a medical blood pressure monitor) will be used for verification, and the data will be marked as high-risk data to trigger an alert.

[0061] As an example, two-dimensional conflict adjudication can be achieved by calculating the confidence level of the conflicting data; the higher the confidence level, the more reliable the data. The confidence level of a two-dimensional conflict adjudication can satisfy the following formula: ; in, For the confidence level of the two-dimensional conflict resolution, This represents the mean of the reference interval for the medical knowledge base. This represents the mean of the individual baseline reference interval. and The weighting coefficients (individual baselines have higher priority) are used for weighting. For example, if Conf ≥ 0.6, the data is considered valid; if Conf < 0.6, the review process is initiated.

[0062] In step 202, subtype stratification and sample augmentation can be performed on elderly patients. Based on a three-dimensional classification standard of acute illness type, underlying medical history, and rehabilitation stage, elderly patients are divided into multiple subtypes, achieving refined classification of the elderly patient population and facilitating differentiated analysis and processing for different characteristic groups. Within the same subtype, homologous sample difference processing is performed on individual health data, calculating the difference between the individual's health data and the subtype's mean data to highlight the individual's deviation characteristics relative to the same population, enhancing the identification of abnormal conditions. Furthermore, when there are niche subtypes with a sample size less than a set number, a transfer learning algorithm is used to fine-tune the model using labeled data from similar subtypes, generating augmented samples for the niche subtypes. This addresses the problems of sparse samples, insufficient data, and difficulty in model training for niche subtypes, improving data completeness and generalization ability.

[0063] Next, multi-scale time decay weight optimization is performed. A time decay function is constructed based on the difference between the data acquisition time and the current time, the decay coefficient, and the indicator type weight. The weight of the target time period data in the time decay function is strengthened based on the weight strengthening rule to determine the dynamic time decay weight of each health data in the multi-source health data. This gives higher weight to recent key data and gradually reduces the weight of long-term data over time. At the same time, the weight allocation is dynamically adjusted in combination with the importance of indicators to highlight high-value time series information.

[0064] Finally, the health data processed by homologous sample difference analysis is integrated with the enhanced samples of niche subtypes to obtain optimized health data. This optimized health data is then fused with corresponding dynamic time decay weights to obtain integrated data. This effectively eliminates conflicts between multi-source data, fills gaps in niche subtype data, and strengthens key time-series information, making the integrated data more accurate, complete, and aligned with individual disease progression patterns. This provides a high-quality data foundation for subsequent feature extraction, risk prediction, and intervention plan generation.

[0065] In step 203, a feature extraction layer can be constructed using a 3-layer CNN (3×3 convolutional kernels, 32 / 64 / 128 neurons) and a 2-layer LSTM (128 / 64 neurons). The convolutional neural network performs local perception and feature convolution operations on the integrated data to extract spatial correlation features showing mutual influence and coupling between multiple health indicators at the same time. A long short-term memory network is used to perform time-series modeling on the integrated data, capturing the dependencies, delays, and trend evolution patterns between health indicators at different times, thus obtaining time-series dependent features (such as a 7-day continuous decline in activity level). Branch 1 of the model is for relapse risk prediction: outputting risk level (low / medium / high) and development speed (fast / medium / slow), with parameters dynamically optimized using QLearning. Branch 2 of the model is for intervention effect prediction: outputting predicted response rates for different intervention programs (medication adjustment, exercise intervention, dietary optimization). (∈[0,1]).

[0066] A dual-output model for CNN-LSTM-QLearning and intervention effect prediction is built based on a feature extraction layer. The dual-output model includes a relapse risk prediction branch and an intervention effect prediction branch, capable of simultaneously outputting both relapse risk and intervention effect prediction results. An improved QLearning reward function is used to dynamically optimize the network parameters of the dual-output model. The QLearning reward function satisfies the following formula: ; in, The total reward value, This is a predictor of recurrence risk. For false alarm rate, The average response rate of the intervention program. , and These are the weighting coefficients. .

[0067] The extracted spatial correlation features and temporal dependency features are input into the parameter-optimized dual-output model to obtain the predicted values ​​of relapse risk and intervention effect, thus achieving an integrated output of risk prediction and effect evaluation.

[0068] By combining CNN and LSTM structures, both spatial correlations and temporal variation patterns among indicators can be captured simultaneously. Introducing QLearning reinforcement learning to dynamically optimize the model significantly improves prediction accuracy, reduces false positive rates, and ensures that the model output matches the clinical intervention response, thereby enhancing the model's practicality and reliability.

[0069] In step 203, a temporal causal discovery algorithm can also be used to mine the complex precursor causal relationships of disease relapse in elderly patients from the integrated data. The complex precursor causal relationship is the sequential triggering, transmission, and evolution relationship among multiple hidden precursors of disease relapse in elderly patients, constructing a temporal causal relationship chain. The starting indicators, target nodes, and ending indicators of the temporal causal relationship chain are identified as potential intervention targets.

[0070] For example, the Temporal Causal Discovery Algorithm (TCDA), based on Granger causality testing and attention mechanisms, can be used to uncover causal chains between precursor indicators (such as a sudden drop in activity → elevated CRP → abnormal heart rate variability). The causal strength can be calculated using the following formula: ; in, For mutual information, measurement indicators and The degree of correlation; For time lag (1-72 hours, dynamically optimized); As an indicator The entropy value; As an indicator Attention weights (initialized based on the medical knowledge base, dynamically optimized by the model).

[0071] The output includes causal chains (such as a sudden drop in activity levels). → CRP elevation (t+12h) → abnormal heart rate (t+24h), causal strength (≥0.6 is strong causation), intervention target (causal chain initiation index).

[0072] Based on the relapse risk prediction value, the temporal causal relationship chain and intervention target are dynamically optimized. The relapse risk prediction value can include the relapse risk level and the risk development speed.

[0073] Risk level calculation can satisfy the following formula: ; in, Risk level, The initial risk score (∈[0,1]) is the output of CNN-LSTM. The total strength of the causal relationship chain (summing the strengths of all strong causal relationships); and These are the weighting coefficients. .

[0074] The rate of development can be determined based on the length and strength of the causal chain, and the rate of development can satisfy a series of formulas: ; in, For the speed of development, The length of the causal chain. For average time lag, ≥0.2 indicates a fast speed, 0.1≤ <0.2 is considered a medium speed. <0.1 indicates a slow speed.

[0075] In one example, if the relapse risk level is high and the risk development speed is within the first speed range, the starting indicator of the time-series causal relationship chain will be used as the core intervention target to achieve early blocking of disease relapse.

[0076] If the relapse risk level is medium and the risk development speed is in the second speed range, then the target node of the temporal causal relationship chain will be used as the main intervention target to achieve precise control over the disease development process.

[0077] If the relapse risk level is low and the risk development rate is in the third speed range, then the terminal indicator of the temporal causal relationship chain will be used as an auxiliary intervention target to achieve stable regulation of the rehabilitation status.

[0078] Among them, the risk development speed of the first speed range is greater than that of the second speed range, and the risk development speed of the second speed range is greater than that of the third speed range.

[0079] By employing time-series causal discovery algorithms, we can elevate traditional correlation analysis to a causal level, accurately identifying key pathways and controllable nodes in disease progression. By combining relapse risk levels with the rate of risk development to categorize and dynamically optimize intervention targets, we can ensure a high degree of alignment between intervention targets and the current level and rate of disease progression. This enhances the targeting, timeliness, and effectiveness of interventions, providing a precise basis for generating subsequent stratified intervention plans.

[0080] In step 204, based on the optimized temporal causal relationship chain and intervention targets, and combined with the relapse risk level and risk development speed, differentiated interventions are implemented for different risk scenarios to generate tiered intervention plans. The tiered intervention plan is a set of graded, categorized, and personalized intervention strategies formulated according to the relapse risk level, risk development speed, and intervention targets. It can be graded, categorized, and time-based according to risk level, development speed, and intervention target type, including intervention methods, intervention intensity, intervention timing, monitoring frequency, etc., enabling precise and personalized interventions for elderly patients in different disease states.

[0081] For example, high risk + rapid pace: prioritize intervention at the starting point of the causal chain (e.g., "sudden drop in activity level → elevated CRP," adjust exercise programs to increase activity level), coupled with urgent follow-up recommendations (seek medical attention within 24 hours). Medium risk + medium pace: jointly intervene at key nodes of the causal chain (e.g., dietary adjustments + inflammation control), increase monitoring frequency (twice daily for core indicators). Low risk + slow pace: mild intervention (e.g., dietary fiber supplementation + 30 minutes of walking daily), with recommendations sent once a week.

[0082] During the intervention process, the implementation rate of the intervention, the rate of achievement of health indicators, and the predicted value of the intervention effect are collected in real time, and the stratified intervention plan is dynamically adjusted based on the collection results. The implementation rate of the intervention is used to quantitatively assess the degree to which the intervention measures are implemented and carried out in the actual scenario. The rate of achievement of health indicators is the proportion of health indicators that return to the normal range or the individual's dynamic baseline after the intervention. The predicted value of the intervention effect is output by the dual-output model and is used to characterize the expected effect of the intervention measures on improving the condition and reducing the risk.

[0083] The dynamic adjustment of the intervention plan can satisfy the following formula: ; in, The original intervention intensity; The intervention implementation rate (0-100%, obtained through feedback from smart devices); , indicating a change in health status (ΔH>0 indicates improvement); This is the predicted value for the intervention response rate. For example, if the implementation rate is low (E<60%), the intervention intensity should be increased; if the health status improves (ΔH>0), the intensity should be decreased; and if the response rate is high (Rint≥0.7), the core intervention content should be maintained.

[0084] The predictive value of the intervention effect on the causal chain assesses the change in the total strength of the causal chain after intervention. ΔC > 0 indicates that the intervention is effective (the causal chain is blocked). The assessment formula can satisfy: ; in, This is the predicted value for the intervention effect. The total strength of the causal chain before intervention; This represents the total strength of the causal chain after intervention (2 weeks). For example, ΔC ≥ 0.2 indicates a significant intervention effect, while ΔC < 0 indicates an ineffective intervention requiring adjustment of the protocol.

[0085] If the implementation rate of the intervention is lower than the set implementation rate, it is determined that the intervention measures have not been effectively implemented, and the intensity of the intervention will be increased accordingly, and the reminder and supervision mechanisms will be strengthened.

[0086] If the rate of health indicator compliance is lower than the set compliance rate and the predicted value of intervention effect is greater than the set effect value, it indicates that the current intervention target or intervention method does not match the patient's actual condition, and the intervention target and intervention method should be adjusted adaptively.

[0087] If the predicted value of the intervention effect is greater than or equal to the set effect value, it indicates that the current intervention has achieved the expected effect. The intensity of the intervention and the monitoring frequency should be reduced accordingly to reduce interference with the patient while ensuring the intervention effect.

[0088] Then, an evaluation system was constructed, comprising multiple assessment indicators including intervention implementation rate, health indicator achievement rate, predicted intervention effect, reduction in relapse risk, and changes in the strength of the causal chain. This evaluation system, composed of multiple intervention-related indicators, is used to comprehensively evaluate the effectiveness and rationality of the intervention program. Each evaluation indicator is assessed at set intervals, a comprehensive evaluation value is calculated, and the effectiveness of the intervention program is determined based on the comprehensive evaluation value. If the comprehensive evaluation value is greater than or equal to the set evaluation value and the intervention program is effective, the optimized intervention program is maintained. Conversely, if the comprehensive evaluation value is less than the set evaluation value, the intervention program is iteratively optimized again based on the temporal causal chain and the intervention target.

[0089] By generating tiered intervention plans based on temporal causal chains and intervention targets, precise intervention can be achieved for latent conditions during the recovery period of acute illnesses in the elderly. Dynamic adjustments are made based on intervention implementation rate, health indicator achievement rate, and predicted intervention effect, forming a closed-loop control mechanism of monitoring-prediction-intervention-feedback-optimization, thus improving the adaptability and feasibility of the intervention plan. By constructing a multi-indicator comprehensive evaluation system, the intervention effect can be objectively and comprehensively evaluated, intervention strategies can be continuously optimized, the risk of relapse during the recovery period of acute illnesses in the elderly can be effectively reduced, and the safety and quality of rehabilitation can be improved.

[0090] This application's embodiments achieve a closed-loop process and system optimization. Through data collection → data integration → AI decision-making → personalized intervention → effect evaluation → feedback optimization (adjusting collection parameters, integration weights, and model parameters), a continuously iterative closed-loop system is formed.

[0091] Dynamic model optimization can be achieved by collecting data from 50 new cases every 3 months, updating model parameters using an incremental learning algorithm (IL-CNN-LSTM), freezing the underlying feature extraction layer, and fine-tuning the attention and fully connected layers (learning rate = 0.0005). Rule base iteration can be achieved by combining clinical feedback and the latest guidelines, updating the anomaly triggering rule base and medical knowledge base benchmarks every 6 months. Privacy protection can be achieved by using federated learning and homomorphic encryption to process sensitive data; only model parameters are transmitted during training, without disclosing the original data.

[0092] Figure 3 This is a schematic diagram of the structure of a monitoring device 300 for monitoring the concealed symptoms of elderly patients during the recovery period of acute illness, provided in an embodiment of this application. Figure 3 As shown, the monitoring device 300 for the concealed condition during the recovery period of acute illness in the elderly may include a trigger module 301, an integration module 302, an optimization module 303, and an evaluation module 304.

[0093] The trigger module 301 is used for a dual-drive data acquisition and scheduling mechanism based on the rehabilitation stage and relapse risk level of elderly patients to acquire health indicator monitoring data. When a health indicator is detected to be outside the baseline range, it triggers the hierarchical linkage acquisition of specific health indicators to obtain multi-source health data.

[0094] The integration module 302 is used to make two-dimensional data conflict resolution on multi-source health data based on medical knowledge base benchmarks and individual dynamic baselines, and to classify patients into subtypes according to acute type, underlying medical history and rehabilitation stage, and to perform cross-subtype transfer enhancement on multi-source health data. It also dynamically adjusts the time decay weights in combination with rehabilitation stage and indicator type to obtain integrated data.

[0095] The optimization module 303 is used to extract the spatial correlation features and temporal dependence features of the integrated data and input them into the preset dual-output model to obtain the relapse risk prediction value and the intervention effect prediction value. The temporal causal discovery algorithm is used to mine the temporal causal relationship chain and intervention target of the compound precursor, and the temporal causal relationship chain and intervention target are optimized based on the relapse risk prediction value.

[0096] The evaluation module 304 is used to generate a stratified intervention plan based on the optimized temporal causal relationship chain and intervention target, and dynamically adjust the stratified intervention plan by combining the intervention implementation rate, the health indicator achievement rate and the predicted value of the intervention effect, and obtain the evaluation results of the stratified intervention plan.

[0097] Among them, the trigger module 301, the integration module 302, the optimization module 303 and the evaluation module 304 can be used to execute steps 201-204 in the embodiment of the above-mentioned monitoring method for hidden conditions during the recovery period of acute diseases in the elderly. For the specific implementation of these modules and more details, please refer to the corresponding method section, which will not be elaborated here.

[0098] This application also provides a computer-readable storage medium storing a program that can be loaded by a processor and executed by any of the monitoring methods for hidden symptoms during the recovery period of acute illness in the elderly according to this application.

[0099] Those skilled in the art will understand that all or part of the functions of the various methods in the above embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, which may include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to achieve the above functions. For example, the program can be stored in the memory of a device, and when the program in the memory is executed by the processor, all or part of the above functions can be achieved. In addition, when all or part of the functions in the above embodiments are implemented by computer programs, the program can also be stored in a server, another computer, disk, optical disk, flash drive, or external hard drive, etc., and can be downloaded or copied to the memory of a local device, or the system of the local device can be updated. When the program in the memory is executed by the processor, all or part of the functions in the above embodiments can be achieved.

[0100] The above examples illustrate this application only to aid understanding and are not intended to limit its scope. Those skilled in the art to which this application pertains can make various simple deductions, modifications, or substitutions based on the ideas presented.

Claims

1. A method for monitoring insidious symptoms during the recovery period of acute illness in the elderly, characterized in that, include: Based on the dual-drive data collection and scheduling mechanism of elderly patients’ rehabilitation stage and relapse risk level, health indicator monitoring data is obtained, and when health indicators are detected to exceed the baseline range, the hierarchical linkage collection of specific health indicators is triggered to obtain multi-source health data. Based on the medical knowledge base benchmark and individual dynamic baseline, the multi-source health data is subject to two-dimensional data conflict adjudication. Patient subtypes are classified according to acute type, underlying medical history and rehabilitation stage. Cross-subtype transfer enhancement is performed on the multi-source health data. The time decay weight is dynamically adjusted in combination with rehabilitation stage and indicator type to obtain integrated data. The spatial correlation features and temporal dependency features of the integrated data are extracted and input into a preset dual-output model to obtain the relapse risk prediction value and the intervention effect prediction value. The temporal causal relationship chain and intervention target of the compound precursor are mined by the temporal causal discovery algorithm, and the temporal causal relationship chain and intervention target are optimized based on the relapse risk prediction value. Based on the optimized temporal causal relationship chain and intervention targets, a tiered intervention plan is generated. The tiered intervention plan is dynamically adjusted by combining the intervention implementation rate, the health indicator achievement rate and the predicted value of the intervention effect, and the evaluation results of the tiered intervention plan are obtained.

2. The method for monitoring latent symptoms during the recovery period of acute illness in the elderly according to claim 1, characterized in that, The dual-drive data acquisition and scheduling mechanism based on the elderly patient's rehabilitation stage and relapse risk level acquires health indicator monitoring data, including: According to clinical standards, the rehabilitation stages of the elderly patients are divided into acute rehabilitation, recovery, and stable phases. Based on the real-time health indicator monitoring data of the elderly patients, the recurrence risk level is output through the initial risk model, which includes low recurrence risk level, medium recurrence risk level and high recurrence risk level. Based on the combination of the rehabilitation stage and the relapse risk level, the collection frequency, priority, and device coordination logic of the health indicators are dynamically adjusted. The collection frequency of the health indicators gradually decreases as the rehabilitation stage progresses and gradually increases as the relapse risk level rises. When the relapse risk level is high, all monitoring devices are activated to work together, and when the risk level is low, unnecessary monitoring devices are turned off.

3. The method for monitoring latent symptoms during the recovery period of acute illness in the elderly according to claim 1, characterized in that, When a health indicator is detected to be outside the baseline range, a tiered, linked collection of specific health indicators is triggered to obtain multi-source health data, including: Integrate typical abnormal patterns of acute diseases in the elderly, formulate rules for linking basic signal abnormalities and specific indicators, and pre-build a target trigger rule base, which includes the mapping relationship between abnormal signals and specific health indicators; When a health indicator exceeds the individual dynamic baseline or the medical knowledge base benchmark, and the duration is greater than or equal to the set duration, it is determined to be a valid abnormal trigger condition. Based on the medical importance coefficient of clinical guidelines, the individual correlation between indicators and previous acute illnesses, and the risk sensitivity coefficient of indicator abnormalities and recurrence risk, a dynamic calculation model for indicator priority is used to determine the collection weight of specific indicators. Based on the valid abnormal basic signals, the corresponding special health indicators are matched from the target trigger rule base, the trigger intensity is divided according to the duration of the abnormality, and an early warning is sent to the medical terminal. Based on the priority of specific indicators, specific health data from multiple monitoring devices are collected synchronously and summarized with basic monitoring data to obtain the multi-source health data.

4. The method for monitoring latent symptoms during the recovery period of acute illness in the elderly according to claim 1, characterized in that, The method of adjudicating data conflicts in the multi-source health data based on medical knowledge base benchmarks and individual dynamic baselines includes: The baseline reference ranges for each health indicator are determined based on the medical knowledge base to construct the medical knowledge base baseline. Based on the historical health data of the elderly patients, the real-time dynamic reference area is determined through a sliding window algorithm to construct the individual dynamic baseline. When the deviation of the same health indicator value obtained from different acquisition devices or data sources exceeds the set deviation range, it is determined to be a multi-source data conflict. Verify whether each set of conflicting data is within the range of the individual dynamic baseline. If only a single conflicting data is within the range of the individual dynamic baseline, then the conflicting data is considered valid data. If all conflicting data are within or outside the individual dynamic baseline, calculate the degree of closeness between each conflicting data and the medical knowledge base benchmark, and take the value with the highest degree of closeness as valid data. If all conflicting data exceed the range of the individual dynamic baseline and the medical knowledge base benchmark, a third-party medical device will be activated to review the data, and the conflicting data will be marked as high-risk data and an early warning will be triggered.

5. The method for monitoring latent symptoms during the recovery period of acute illness in the elderly according to claim 1, characterized in that, The process involves classifying patients into subtypes based on acute illness type, underlying medical history, and recovery stage, and performing cross-subtype transfer enhancement on the multi-source health data. This is combined with dynamically adjusting time decay weights based on recovery stage and indicator type to obtain integrated data, including: Based on the three-dimensional classification criteria of acute type, underlying medical history, and rehabilitation stage, the elderly patients were divided into multiple subtypes. For individual health data within the same subtype, homogeneous sample difference processing is performed to calculate the difference between the individual health data and the subtype data mean. When there is a niche subtype with a sample size less than a set number, a transfer learning algorithm is used to fine-tune the model using labeled data of similar subtypes to generate enhanced samples for the niche subtype. A time decay function is constructed based on the difference between the data acquisition time and the current time, the decay coefficient, and the indicator type weight. The target time period data weight of the time decay function is strengthened based on the weight strengthening rule to determine the dynamic time decay weight of each health data in the multi-source health data. The health data processed by the difference of homologous samples is integrated with the enhanced samples of the minority subtype to form the optimized health data. The optimized health data is then fused with the corresponding dynamic time decay weight to obtain the integrated data.

6. The method for monitoring latent symptoms during the recovery period of acute illness in the elderly according to claim 1, characterized in that, The spatial correlation features and temporal dependency features of the integrated data are extracted and input into a preset dual-output model to obtain relapse risk prediction values ​​and intervention effect prediction values, including: A feature extraction layer is constructed using a 3-layer convolutional neural network and a 2-layer long short-term memory network. The spatial correlation features of the integrated data are extracted by the convolutional neural network, and the temporal dependency features of the integrated data are captured by the long short-term memory network. Based on the feature extraction layer, a dual-output model for CNN-LSTM-QLearning and intervention effect prediction is constructed. The dual-output model includes a relapse risk prediction branch and an intervention effect prediction branch. The parameters of the dual-output model are dynamically optimized using an improved QLearning reward function; The extracted spatial correlation features and temporal dependency features are input into the parameter-optimized dual-output model to obtain the relapse risk prediction value and the intervention effect prediction value. The QLearning reward function satisfies the following formula: ； in, The total reward value, The recurrence risk prediction value is... For false alarm rate, The average response rate of the intervention program. , and These are the weighting coefficients.

7. The method for monitoring latent symptoms during the recovery period of acute illness in the elderly according to claim 1, characterized in that, The step of employing a temporal causal discovery algorithm to mine the temporal causal relationship chain and intervention targets of compound precursors, and optimizing the temporal causal relationship chain and intervention targets based on the recurrence risk prediction value, includes: Using a time-series causal discovery algorithm, the complex precursor causal relationships of disease relapse in elderly patients are mined from the integrated data, and the starting indicators, target nodes, and ending indicators of the time-series causal relationship chain are identified as potential intervention targets. Based on the relapse risk prediction value, the time-series causal relationship chain and the intervention target are dynamically optimized. The relapse risk prediction value includes the relapse risk level and the risk development speed. The process of dynamically optimizing the temporal causal chain and the intervention target based on the predicted relapse risk value includes: If the recurrence risk level is high and the risk development speed is within the first speed range, then the starting indicator of the time-series causal relationship chain will be used as the core intervention target. If the recurrence risk level is medium and the risk development speed is within the second speed range, then the target node of the time-series causal relationship chain will be used as the main intervention target. If the recurrence risk level is low and the risk development speed is in the third speed range, then the terminal indicator of the time-series causal relationship chain will be used as an auxiliary intervention target. The risk development rate of the first speed range is greater than that of the second speed range, and the risk development rate of the second speed range is greater than that of the third speed range.

8. The method for monitoring latent symptoms during the recovery period of acute illness in the elderly according to claim 7, characterized in that, Based on the optimized temporal causal chain and intervention targets, a tiered intervention plan is generated. Combining intervention implementation rate, health indicator achievement rate, and predicted intervention effect values, the tiered intervention plan is dynamically adjusted, and the evaluation results of the tiered intervention plan are obtained, including: Based on the optimized temporal causal relationship chain and the intervention target, combined with the relapse risk level and the risk development speed, differentiated interventions are implemented for different risk scenarios to generate the stratified intervention plan; Real-time data collection of the intervention implementation rate, the health indicator compliance rate, and the predicted intervention effect; If the intervention implementation rate is lower than the set implementation rate, the intervention intensity will be increased; If the health indicator compliance rate is lower than the set compliance rate and the predicted value of the intervention effect is greater than the set effect value, then the intervention target and intervention method shall be adjusted. If the predicted value of the intervention effect is greater than or equal to the set effect value, then the intervention intensity and monitoring frequency should be reduced. Construct an evaluation system that includes multiple evaluation indicators such as the degree of implementation of the intervention, the rate of achievement of the health indicators, the predicted value of the intervention effect, the degree of reduction in relapse risk, and the change in the strength of the causal chain. At set intervals, each evaluation indicator is assessed, a comprehensive evaluation value is calculated, and the effectiveness of the intervention plan is determined. If the overall evaluation value is greater than or equal to the set evaluation value and the intervention plan is effective, then the optimized intervention plan is maintained.

9. A monitoring device for the concealed symptoms of elderly patients during the recovery period from acute illness, characterized in that, include: The triggering module is used to acquire health indicator monitoring data based on the dual-drive collection and scheduling mechanism of the elderly patient's rehabilitation stage and relapse risk level. When the health indicator is detected to be outside the baseline range, it triggers the hierarchical linkage collection of specific health indicators to obtain multi-source health data. The integration module is used to perform two-dimensional data conflict adjudication on the multi-source health data based on the medical knowledge base benchmark and individual dynamic baseline, and to classify patients into subtypes according to acute type, basic medical history and rehabilitation stage, and to perform cross-subtype transfer enhancement on the multi-source health data. The time decay weight is dynamically adjusted in combination with rehabilitation stage and indicator type to obtain integrated data. The optimization module is used to extract the spatial correlation features and temporal dependency features of the integrated data and input them into a preset dual-output model to obtain the relapse risk prediction value and the intervention effect prediction value. The temporal causal discovery algorithm is used to mine the temporal causal relationship chain and intervention target of the compound precursor, and the temporal causal relationship chain and intervention target are optimized based on the relapse risk prediction value. The evaluation module is used to generate a stratified intervention plan based on the optimized temporal causal relationship chain and intervention targets, dynamically adjust the stratified intervention plan by combining the intervention implementation rate, the health indicator achievement rate and the predicted value of the intervention effect, and obtain the evaluation results of the stratified intervention plan.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that can be loaded by a processor and executed as described in any one of claims 1 to 8 for monitoring the concealed condition during the recovery period of acute illness in the elderly.

Images