Big data-based scientific research data analysis method and system for cognitive impairment in the elderly

Abstract

The application provides a kind of old cognitive impairment scientific research data analysis method and system based on big data, it is related to big data technical field, first, obtain clinical diagnosis and treatment record, neuroimaging examination, cognitive function assessment and follow-up tracking and so on Multiple types of old cognitive impairment scientific research data are integrated into scientific research big data set;Then dynamically associate intervention-related information and data change information therein, build intervention feedback link containing multiple nodes;From the intervention feedback link, the non-explicit association of different nodes in the time dimension is mined, and the implicit time sequence association set is formed;Based on the implicit time sequence association set, the link of multiple research objects is hierarchically aggregated, the group commonness rule and the individual difference rule are extracted, and the group individual rule set is formed;Finally, according to the group individual rule set, the adaptive relationship and the condition are arranged, and the scientific research data analysis result report is generated, which provides effective reference scientific research content for the prevention and treatment of old cognitive impairment.

Description

Technical Field

[0001] This invention relates to the field of big data technology, and more specifically, to a method and system for analyzing scientific research data on cognitive impairment in the elderly based on big data. Background Technology

[0002] In the field of cognitive impairment research in the elderly, with the continuous advancement of medical technology and the in-depth development of scientific research, a large amount of diverse research data has been accumulated, covering multiple aspects such as clinical diagnosis and treatment records, neuroimaging examinations, cognitive function assessments, and follow-up tracking. However, current methods for analyzing the aforementioned research data on cognitive impairment in the elderly have many limitations.

[0003] Traditional data analysis methods often process various types of data in isolation, lacking in-depth exploration of the intrinsic connections between different data types. When analyzing the impact of interventions on cognitive impairment in the elderly, they only focus on direct data changes after the intervention, failing to dynamically link intervention-related information with data change information to construct a complete intervention feedback mechanism. Furthermore, existing methods struggle to effectively capture the potential correlations between different data units over time, resulting in an inability to comprehensively and deeply understand the development process of cognitive impairment in the elderly and the long-term effects of interventions. In addition, when generalizing from individual studies to group patterns, the lack of scientifically effective methods to distinguish between common patterns and individual differences limits the application of research results and hinders precise guidance for the prevention and treatment of cognitive impairment in the elderly. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for analyzing research data on cognitive impairment in the elderly based on big data, the method comprising:

[0005] Acquire research data on various types of cognitive impairment in the elderly and integrate them to form a research big data set, which includes clinical diagnosis and treatment record data, neuroimaging examination data, cognitive function assessment data, and follow-up tracking data.

[0006] The intervention-related information and data change information in the scientific research big data set are dynamically correlated to construct an intervention feedback link with multiple nodes. Each node in the intervention feedback link corresponds to a type of data unit and the association attribute between the data unit and the intervention measures.

[0007] The non-explicit correlations between different nodes in the time dimension are mined from the intervention feedback link to form a set of implicit temporal correlations, which contains the potential correlations between different data units that evolve over time.

[0008] Based on the implicit temporal correlation set, the intervention feedback links of multiple research subjects are hierarchically aggregated to extract common patterns applicable to the group and individual difference patterns that exist only in a single research subject, forming a set of group-individual patterns;

[0009] Based on the set of individual patterns in the group, the matching relationship between intervention measures and data changes, as well as individual matching conditions, are organized to generate a research data analysis report on cognitive impairment in the elderly.

[0010] Furthermore, embodiments of the present invention also provide a big data-based research data analysis system for cognitive impairment in the elderly, characterized in that it includes:

[0011] A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the above-described big data-based research data analysis method for cognitive impairment in the elderly by executing the machine-executable instructions.

[0012] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, the processor of a computer device reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the computer device to execute the above-mentioned method for analyzing scientific research data on cognitive impairment in the elderly based on big data.

[0013] Based on the above, a large-scale research dataset of cognitive impairment in the elderly is formed by acquiring and integrating various types of research data. Then, intervention-related information and data change information in the research dataset are dynamically correlated and an intervention feedback chain is constructed. This can present the real-time interactive relationship between intervention measures and data changes. Next, the implicit temporal correlations of different nodes in the intervention feedback chain are mined to form a set of implicit temporal correlations, which reveals the potential correlations between different data units over time. Based on the implicit temporal correlation set, the intervention feedback chains of multiple research subjects are hierarchically aggregated to extract common patterns and individual differences applicable to the group, forming a set of group-individual patterns. This can grasp the general characteristics of cognitive impairment in the elderly at the group level while also paying attention to the particularities of individuals. Finally, the adaptation relationship between intervention measures and data changes and individual adaptation conditions are sorted out according to the set of group-individual patterns to generate an analysis report, thereby improving the reliability of research data analysis on cognitive impairment in the elderly. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the execution flow of the research data analysis method for cognitive impairment in the elderly based on big data provided in the embodiments of the present invention.

[0015] Figure 2This is a schematic diagram of exemplary hardware and software components of the big data-based scientific research data analysis system for cognitive impairment in the elderly provided in an embodiment of the present invention. Detailed Implementation

[0016] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a research data analysis method for cognitive impairment in the elderly based on big data, provided in one embodiment of the present invention. The following is a detailed description of this research data analysis method for cognitive impairment in the elderly based on big data.

[0017] Step S110: Acquire research data on various types of cognitive impairment in the elderly and integrate them to form a research big data set, which includes clinical diagnosis and treatment record data, neuroimaging examination data, cognitive function assessment data, and follow-up tracking data.

[0018] In this embodiment, the various types of data collected from different medical institutions or research centers are first standardized in terms of format. Clinical diagnosis and treatment record data may come from different electronic medical record systems, and their data formats differ. They need to be uniformly converted into structured document formats, such as XML or JSON, for subsequent data processing and analysis. Neuroimaging examination data typically includes magnetic resonance imaging data, computed tomography (CT) scan data, etc., and these data need to be stored and integrated according to the standard format of medical image storage and transmission systems.

[0019] Cognitive function assessment data are generally scale scores, such as Mini-Mental State Examination scores and Montreal Cognitive Assessment scores. These need to be converted into numerical data and linked to the corresponding research subjects. Follow-up tracking data includes patients' daily activity records, medication records, etc., and must be arranged in chronological order to ensure the timeliness of the data.

[0020] After standardizing the format, a unified data index is established to link different types of data through unique research object identifiers, thus forming a large-scale scientific research dataset. Each data entry in this dataset contains basic information about the research object (already anonymized), corresponding examination and evaluation data, and timestamp information to facilitate subsequent analysis and mining of the data over time.

[0021] Step S120: Dynamically correlate intervention-related information and data change information in the scientific research big data set to construct an intervention feedback link with multiple nodes. Each node in the intervention feedback link corresponds to a type of data unit and the association attribute between the data unit and the intervention measures.

[0022] Step S121: Extract intervention-related information from the clinical diagnosis and treatment record data of the scientific research big data set, and break it down into intervention type nodes, intervention implementation nodes, intervention adjustment nodes, and intervention termination nodes. Each node records the specific content of the corresponding intervention link.

[0023] In this embodiment, intervention-related information refers to a complete set of information extracted from clinical diagnosis and treatment records in the big data of research on cognitive impairment in the elderly, directly related to the diagnostic and treatment interventions implemented for elderly patients with cognitive impairment. Its core function is to reflect the type of intervention, implementation details, adjustment process, and termination status. The big data of research on cognitive impairment in the elderly refers to a multi-dimensional, time-series set of clinical and research data collected and integrated through standardized processes in the field of scientific research on cognitive impairment in the elderly in order to systematically explore the patterns of disease progression and the effects of intervention measures. Its core components include clinical diagnosis and treatment record data from medical institutions' electronic medical record systems, including structured text information directly related to clinical interventions such as diagnostic conclusions, treatment plans, and medication records; neuroimaging examination data obtained through equipment such as magnetic resonance imaging and computed tomography, reflecting quantitative imaging indicators of brain structure and function, such as the volume of specific brain regions and gray matter density; cognitive function assessment data obtained through periodic testing using standardized scales such as the Mini-Mental State Examination and the Montreal Cognitive Assessment, which objectively assess the status of cognitive domains such as memory and executive function with quantitative scores; and follow-up tracking data collected through regular follow-ups during non-hospitalization periods, covering dynamic changes in patients' daily living abilities, behavioral symptoms, and medication adherence.

[0024] The intervention-related information specifically includes the following:

[0025] Intervention type information: refers to the specific category and core attributes of the diagnostic and treatment interventions adopted for elderly patients with cognitive impairment, including but not limited to drug intervention types (such as specific drug types and pharmacological properties of cholinesterase inhibitors, NMDA receptor antagonists, etc.), cognitive training intervention types (such as specific training directions such as memory training, attention training, executive function training, etc.), rehabilitation therapy intervention types (such as specific rehabilitation methods such as motor rehabilitation, language rehabilitation, and activities of daily living rehabilitation, etc.), and behavioral intervention types (such as specific intervention methods such as adjusting work and rest, dietary guidance, and psychological counseling, etc.). Each intervention type clearly corresponds to the core intervention direction recorded in the clinical diagnosis and treatment records.

[0026] Intervention Implementation Information: This refers to the specific operational details and basic parameters when the intervention is formally implemented, including the specific implementation method of the intervention (such as oral or injectable administration of drugs, and offline centralized training or online home training for cognitive training), implementation intensity (such as daily dosage and frequency of drug administration, and duration of a single training session and number of training sessions per week for cognitive training), implementation start time (accurate to the intervention start date and specific time recorded in the clinical diagnosis and treatment record), and implementation scenario (such as implementation in a hospital, at home, or in a rehabilitation institution). All information is directly derived from the intervention implementation record entries in the clinical diagnosis and treatment record.

[0027] Intervention adjustment information: This refers to the details of modifications made to the original intervention plan based on patient feedback (such as changes in symptoms, fluctuations in test results, etc.) during the intervention implementation process. This includes adjustment trigger conditions (such as reasons for adjustment recorded in clinical records, such as a decline in cognitive assessment scores, the occurrence of drug side effects, changes in neuroimaging indicators, etc.), adjustment content (such as specific modifications such as increases or decreases in drug dosage, changes in cognitive training methods, adjustments in the intensity of rehabilitation treatment, etc.), adjustment time (the date and time of adjustment implementation recorded in the clinical diagnosis and treatment records), and adjustment basis (such as the physician's judgment based on clinical guidelines, the patient's individual tolerance, abnormal test indicators, etc.).

[0028] Intervention termination information: refers to relevant information when the intervention is stopped, including the reason for termination (such as the intervention achieving the expected effect, the patient experiencing intolerable side effects, the patient's condition progressing to the point of ineffectiveness of the intervention, the patient voluntarily terminating, etc., as recorded in the clinical records), the termination time (the date and time of intervention cessation recorded in the clinical records), the intervention status at the time of termination (such as the last intervention execution parameters such as the drug dosage and training progress at the time of termination), and the follow-up plan after termination (such as whether regular follow-up examinations are required after termination, and the follow-up examination items, etc.).

[0029] When extracting intervention-related information from clinical medical records, the first step is to parse the data. Clinical medical records typically contain unstructured or semi-structured text information such as doctors' diagnoses, treatment plans, and medication records. Natural language processing (NLP) techniques are used to segment, tag, and name entity recognition of this text information to identify entities and events related to the intervention. For example, from the text "A patient was admitted to the hospital due to memory loss and given donepezil tablets orally, with an initial dose of 5 mg daily, adjusted to 10 mg daily after two weeks, and discontinued after three months of continuous use," we can extract the intervention type as "drug therapy (donepezil tablets)." The specific details of the intervention implementation node include the drug name, initial dose, method of administration (oral), and start time. The specific details of the intervention adjustment node include the adjusted dose (10 mg daily) and adjustment time (after two weeks of use). The specific details of the intervention termination node include the discontinuation time (after three months of continuous use). This extracted information is stored in corresponding nodes, each containing a unique node identifier to facilitate the association between nodes when constructing the intervention feedback loop.

[0030] Step S122: Extract data change information from the neuroimaging examination data, cognitive function assessment data and follow-up tracking data of the scientific research big data set, and form multiple nodes respectively. The multiple nodes include neuroimaging nodes, cognitive assessment nodes and follow-up record nodes. Each node records the status change content of the corresponding data type.

[0031] In this embodiment, the data change information refers to a set of information extracted from neuroimaging examination data, cognitive function assessment data, and follow-up tracking data in the big data of scientific research on cognitive impairment in the elderly. This information reflects the evolution of the patient's physiological state, cognitive function, and living status at different time points. Its function is to capture the dynamic changes in the patient's state by comparing similar data at different time points.

[0032] The data change information is specifically categorized by data source as follows:

[0033] Information on changes based on neuroimaging data: This refers to information reflecting the evolution of brain structure and function by comparing neuroimaging data (such as magnetic resonance imaging data, computed tomography data, etc.) at different time points of the same patient. This includes, but is not limited to, changes in the volume of specific brain regions (such as the increase, decrease, or stabilization of the volume of brain regions related to cognitive function, such as the hippocampus and medial temporal lobe), changes in gray matter / white matter density (such as the increase, decrease, and changes in the distribution range of gray matter density), changes in brain metabolic levels (such as the increase or decrease in metabolic activity in specific brain regions), and changes in cerebral blood flow perfusion (such as the evolution of blood supply status in related brain regions). All information on changes is obtained through quantitative analysis and qualitative description of imaging data at different time points.

[0034] Information on changes in cognitive function assessment data: This refers to information reflecting the evolution of cognitive abilities extracted by comparing the cognitive function scale assessment results of the same patient at different time points. This includes, but is not limited to, changes in memory function (such as increases, decreases, or stabilization of memory dimension scores in the Mini-Mental State Examination), changes in attention (such as the evolution of scores in the Digit Span Test), changes in executive function (such as decreases or increases in completion time in the Connecting Test), changes in language function (such as changes in scores in the Naming Test and Fluency Test), and changes in visuospatial ability (such as the evolution of scores in the Block Test). The information on these changes is directly reflected through scale scores and qualitative descriptions by the assessing physician.

[0035] Information on changes based on follow-up tracking data: This refers to information reflecting daily living abilities, mental and behavioral status, and disease progression extracted by comparing follow-up records of the same patient at different time points. This includes, but is not limited to, changes in daily living self-care abilities (such as the evolution of the degree to which basic daily living behaviors such as dressing, washing, and eating are completed independently), changes in mental and behavioral symptoms (such as the appearance, disappearance, or changes in the severity of symptoms such as hallucinations, delusions, anxiety, and depression), changes in the rate of disease progression (such as the time interval between symptom exacerbations and the duration of disease stability), changes in medication adherence (such as the evolution of the patient's adherence to medication as recorded in the follow-up records), and changes in social interaction abilities (such as the evolution of the frequency and quality of communication with family members and others). All information on changes comes from objective descriptions and quantitative assessment items in the follow-up records.

[0036] For neuroimaging data, comparative analysis of images from different time points is necessary to extract information on changes in state. For example, medical imaging analysis software can be used to process a patient's MRI data, measuring indicators such as changes in volume and gray matter density in specific brain regions. Changes in these indicators are recorded in neuroimaging nodes, with each node corresponding to one imaging examination and including examination time, measurement indicator name, and indicator value changes. Extracting information on changes in state from cognitive function assessment data is relatively straightforward. Scale scores at different time points are compared, and the changes in score values ​​and rates are calculated. This information is recorded in cognitive assessment nodes, including assessment time, scale name, and score changes. Information on changes in state from follow-up data is more complex and needs to be extracted from textual information such as the patient's daily activity records and symptom descriptions. For example, from the follow-up record "The patient's family recently reported a decline in their daily living self-care ability, and they easily get lost after going out," information on decreased self-care ability and reduced spatial orientation can be extracted. This information is recorded in the follow-up record node, including follow-up time, description of state changes, and related symptoms.

[0037] Step S123: Determine the temporal relationship between each intervention node and data node. Based on the time information of the intervention implementation node, forward association is performed on the neuroimaging nodes, cognitive assessment nodes, and follow-up record nodes within the preset time period, which are marked as pre-intervention nodes. Forward association is performed on the neuroimaging nodes, cognitive assessment nodes, and follow-up record nodes from the intervention implementation to the intervention termination period, which are marked as post-intervention nodes.

[0038] First, the specific time of intervention implementation is extracted from the intervention implementation node, such as a specific date (year, month, day). The length of the preset time period is determined based on the characteristics of different data types and the needs of clinical research. For neuroimaging data, due to its high cost and relatively long intervals between examinations, the preset time period can be set to three to six months before intervention implementation. For cognitive function assessment data, which has a relatively high examination frequency, the preset time period can be set to one to three months before intervention implementation. The preset time period for follow-up data can be set to two weeks to one month before intervention implementation. Based on the preset time period, the neuroimaging data, cognitive function assessment data, and follow-up data of the research subjects corresponding to the intervention implementation node within the preset time period are searched in the scientific research big data set, and the corresponding nodes are marked as pre-intervention nodes. For example, if the pre-implementation date is March 1st of a certain year, and the preset time period is three months before intervention implementation, then the neuroimaging nodes, cognitive assessment nodes, and follow-up record nodes of the research subjects between January 1st and March 1st of that year are searched, and these nodes are marked as pre-intervention nodes. To determine the post-intervention nodes, the intervention implementation time is used as the starting point and the intervention termination time as the ending point. All neuroimaging nodes, cognitive assessment nodes, and follow-up record nodes within this time period are identified and marked as post-intervention nodes. For example, if the intervention implementation time is March 1st of a certain year and the intervention termination time is June 1st of the same year, then the aforementioned data nodes of the research subject between March 1st and June 1st are marked as post-intervention nodes.

[0039] Step S124: Add association attributes to each node. The association attributes of the pre-intervention node include a description of the degree of influence of the data status of the pre-intervention node on the selection of subsequent intervention measures. The association attributes of the post-intervention node include a description of the response time of the data changes of the post-intervention node and the implementation of intervention measures. The association attributes of the intervention node include a description of the expected direction of the intervention content of the intervention node on the data node.

[0040] Step S1241: For the pre-intervention node, analyze the correspondence between the data status of the pre-intervention node and the intervention type in the subsequent intervention type node, calculate the frequency ratio of selecting a specific intervention under the same data status, and determine the degree of influence of the data status on the selection of intervention based on the frequency ratio. The degree of influence includes three categories: dominant choice, auxiliary reference, and no significant influence.

[0041] When analyzing the correspondence between pre-intervention node data states and intervention types, a large amount of historical research data was first collected. This data included the types of interventions selected under different pre-intervention node data states. For each pre-intervention node data state, the number of times each type of intervention was selected was counted. For example, for a specific cognitive function assessment score range (data state), the number of times different intervention types such as drug therapy, cognitive training, and rehabilitation therapy were selected within that score range was counted. Then, the frequency percentage of a specific intervention was calculated, which is the number of times that intervention was selected divided by the total number of intervention selections under that data state. The degree of influence was described based on the frequency percentage: if the frequency percentage was greater than or equal to 70%, it was determined as "dominant choice"; if the frequency percentage was between 30% and 70%, it was determined as "secondary reference"; and if the frequency percentage was less than 30%, it was determined as "no significant influence". The determined degree of influence was then added to the association attributes of the pre-intervention node.

[0042] Step S1242: For the post-intervention node, extract the start time of the data change of the post-intervention node and the time information of the corresponding intervention implementation node or intervention adjustment node, calculate the time interval between the two, and determine the response timeliness description based on the time interval. The response timeliness description includes four categories: immediate response, short-term response, medium-term response, and long-term response.

[0043] The start time of data change at the post-intervention node refers to the point in time when data changes begin, such as the time when cognitive assessment scores begin to rise or fall, or the time when neuroimaging indicators begin to change. This start time is extracted from the post-intervention node, and the implementation or adjustment time is extracted from the corresponding intervention implementation or adjustment node. The time interval between the two is calculated as the start time of data change minus the intervention implementation or adjustment time. The response timeliness description is determined based on the length of the time interval: if the time interval is less than or equal to one week, it is a "short-term response"; if the time interval is greater than one week but less than or equal to one month, it is a "medium-term response"; if the time interval is greater than one month but less than or equal to three months, it is a "long-term response". The determined response timeliness description is added to the associated attributes of the post-intervention node.

[0044] Step S1243: For intervention type nodes, based on research consensus in the field of cognitive impairment in the elderly, determine the expected direction of the intervention type on the changes in data of neuroimaging nodes, cognitive assessment nodes and follow-up record nodes. The expected direction of the impact is described in four categories: promoting positive changes, inhibiting negative changes, maintaining a stable state, and no clear expectation.

[0045] Research consensus in the field of cognitive impairment in the elderly can be obtained by reviewing authoritative medical literature, clinical guidelines, and expert consensus documents. For each type of intervention, such as cholinesterase inhibitors in drug therapy, the expected direction of their impact on data changes at neuroimaging nodes (e.g., changes in brain volume, changes in brain metabolic rate), cognitive assessment nodes (e.g., changes in cognitive scale scores), and follow-up recording nodes (e.g., changes in daily living abilities, changes in psychosocial symptoms) should be determined based on research consensus. For example, cholinesterase inhibitors are believed to improve cognitive function, therefore their expected impact on changes in cognitive assessment node data is "promoting positive changes"; for neuroimaging nodes, it may be expected to slow the progression of brain atrophy, i.e., "inhibiting negative changes"; for daily living abilities in follow-up recording nodes, it may be expected to "maintain a stable state." If the impact of a certain type of intervention on certain data nodes is not yet clearly concluded in existing studies, the expected direction of impact is described as "no clear expectation." The above description of the expected direction of impact is added to the association attributes of the intervention type node.

[0046] Step S1244: For the intervention implementation node, combined with the expected impact direction of the intervention type node, supplement the expected correlation between the intensity of the intervention and the magnitude of data change. The expected correlation includes three categories: increased intensity may increase the magnitude of change, the intensity of the intervention may exceed the threshold and have a reverse effect, and the intensity of the intervention can be stabilized if it is maintained within a specific range.

[0047] The intensity of intervention implementation can include drug dosage, duration and frequency of cognitive training, and intensity of rehabilitation therapy. The analysis is based on the expected impact direction of the intervention type node, combined with the data changes under different implementation intensities. For example, for a drug treatment intervention type with the expected impact direction of "promoting positive change," when the implementation intensity (dosage) increases within a certain range, the magnitude of data change (such as the increase in cognitive assessment scores) may also increase. In this case, the associated expectation description is "increased intensity may enhance the magnitude of change." If research shows that when the drug dosage exceeds a certain threshold, side effects may occur, leading to a decrease in the magnitude of data change or even a reverse change (such as cognitive function deterioration). In this case, the associated expectation description is "intervention implementation intensity exceeding the threshold may have a reverse effect." For some interventions, when the implementation intensity is maintained within a specific range, the magnitude of data change can remain relatively stable without significant fluctuations. In this case, the associated expectation description is "maintaining the intervention implementation intensity within a specific range can stabilize the effect." These associated expectation descriptions are added to the association attributes of the intervention implementation node.

[0048] Step S1245: For the intervention adjustment node, analyze the correlation between the adjustment content and the data changes of the subsequent intervention post-node, and determine the correction direction description of the adjustment content on the data change. The correction direction description includes three categories: strengthening positive change, mitigating negative change, and optimizing change stability.

[0049] The adjustments made at intervention adjustment nodes may include increasing or decreasing drug dosage, changing treatment plans, and adjusting training intensity. The changes in data at subsequent intervention nodes after these adjustments are analyzed. For example, if during the intervention, the rate of increase in cognitive assessment scores gradually slows (positive change but at a slower pace), and subsequent adjustments to the intervention, such as increasing the frequency of cognitive training, lead to a renewed acceleration in the rate of increase in cognitive assessment scores, the correction direction for the data change in this case is described as "strengthening the positive change." If, after the intervention, the data shows a reverse change, such as neuroimaging indicators showing an accelerated rate of brain atrophy, and subsequent adjustments to the intervention, such as changing the type of medication, slow down the rate of brain atrophy, the correction direction is described as "alleviating the reverse change." If the data fluctuation is large and unstable, adjusting the implementation method of the intervention, such as changing centralized cognitive training to distributed training, reduces the fluctuation of the data fluctuation and makes it more stable, then the correction direction is described as "optimizing the stability of the change." The determined correction direction descriptions are added to the association attributes of the intervention adjustment nodes.

[0050] Step S1246: For the intervention termination node, calculate the maintenance status of the data changes of the post-intervention nodes after the intervention termination node, and determine the duration description of the intervention effect. The duration description includes three categories: rapid decline of effect, short-term maintenance of effect, and long-term maintenance of effect.

[0051] Even after the intervention termination point, follow-up and data collection of the research subjects are still necessary for a period of time to observe the maintenance of the intervention effect. The calculation involves determining the length of time after the intervention termination point that the data changes at the post-intervention milestones remain at the level at the intervention termination point. For example, if the cognitive assessment score at the intervention termination point is a certain value, cognitive assessments are conducted again at the first, second, and third months after the intervention termination to observe whether the score remains at that value or fluctuates within an acceptable range. If the data change significantly returns to the pre-intervention level or shows a reverse change within one month after the intervention termination, the duration of effectiveness is described as "rapid decline of effect"; if the data change can be maintained for one to three months, it is described as "short-term maintenance of effect"; if it can be maintained for more than three months, it is described as "long-term maintenance of effect". The determined duration of effectiveness description is added to the associated attributes of the intervention termination point.

[0052] Step S1247: Bind the above-mentioned association attribute descriptions to the corresponding nodes so that the association attributes of each node reflect the association characteristics and influence relationships between the node and other nodes.

[0053] After determining the association attribute descriptions for various types of nodes, these descriptions are bound to the corresponding nodes using node identifiers. In the data storage structure, a dedicated field is set up for each node to store the association attribute description information. For example, in a relational database, an association attribute field can be added to the node table, storing various association attribute descriptions in text format; in a graph database, attribute key-value pairs can be added to nodes, with the key being the association attribute type and the value being the corresponding association attribute description. In this way, when subsequently analyzing and mining the intervention feedback chain, the association attribute description of each node can be quickly obtained through the node identifier, thereby understanding the association characteristics and influence relationships between nodes.

[0054] Step S125: Establish dynamic connection relationships between nodes. Intervention type nodes are connected to intervention implementation nodes and intervention pre-intervention nodes respectively. Intervention implementation nodes are connected to intervention adjustment nodes and intervention post-intervention nodes respectively. Intervention adjustment nodes are connected to intervention post-intervention nodes in subsequent time periods. Intervention termination nodes are connected to the final state node of intervention post-intervention nodes.

[0055] When establishing dynamic connections between nodes, a directed graph is constructed based on node identifiers to represent these connections. The intervention type node is the starting point of the intervention, pointing to the corresponding intervention implementation node and representing the specific implementation content under that intervention type. Simultaneously, the intervention type node also points to the pre-intervention node, as the data state of the pre-intervention node influences the choice of intervention type. The intervention implementation node is the execution point of the intervention, pointing to the intervention adjustment node and representing adjustments that may be made during the intervention implementation process. It also points to the post-intervention node, representing the impact of the intervention on data changes. The intervention adjustment node points to the post-intervention node in subsequent time periods, because the adjusted intervention will have new impacts on subsequent data changes. The intervention termination node points to the final state node of the post-intervention node, which is the corresponding post-intervention node at the time of intervention termination, reflecting the final effect of the intervention. For example, an intervention type node (drug treatment) is connected to the corresponding intervention implementation node (drug A, dosage B, time C) and pre-intervention nodes (pre-intervention cognitive assessment node, neuroimaging node, etc.); the intervention implementation node is connected to the intervention adjustment node (dosage adjusted to D, time E) and post-intervention nodes (first cognitive assessment node, first neuroimaging node, etc. after intervention implementation); the intervention adjustment node is connected to subsequent post-intervention nodes (adjusted cognitive assessment node, neuroimaging node, etc.); and the intervention termination node is connected to the final state node of the post-intervention nodes (cognitive assessment node, neuroimaging node, etc. at the time of intervention termination). Through these dynamic connections, a complete intervention feedback loop is formed, clearly demonstrating the entire process of intervention measures from selection, implementation, adjustment to termination, and their impact on data changes.

[0056] Step S126: Through the synergistic effect of node connection relationships and associated attributes, each node in the intervention feedback link can be traced back to the corresponding intervention node or data node, and each connection is accompanied by a clear associated attribute description, thereby constructing the intervention feedback link.

[0057] In constructing the intervention feedback chain, the connections between nodes are established based on the aforementioned dynamic connection rules, and each connection includes a description of the related node's attributes. For example, the connection between the intervention type node and the intervention implementation node includes a description of the expected impact direction of the intervention type node and an expected association description of the intervention implementation node. These descriptions explain the selection criteria for the intervention type and the relationship between the implementation intensity and the magnitude of data change. The connection between the intervention implementation node and the post-intervention node includes a description of the post-intervention node's response timeliness, indicating the speed at which data changes respond to the intervention measures. Through this method, each node in the intervention feedback chain can be traced back to its related intervention or data nodes through connection relationships, and the specific attributes of their relationships can be understood. For example, when examining a post-intervention node, the corresponding intervention implementation node and intervention adjustment node can be found through connection relationships, thus understanding which intervention measure caused the data change, and the impact of the intervention's implementation intensity and adjustment on the data change. This traceability and clear attribute descriptions enable the intervention feedback chain to accurately reflect the dynamic relationship between intervention measures and data changes.

[0058] Step S130: Mine the non-explicit correlations of different nodes in the time dimension from the intervention feedback link to form a set of implicit time-series correlations, which contains the potential correlations between different data units that evolve over time.

[0059] Step S131: Extract the time information and status information of all nodes in the intervention feedback chain. The time information is the collection or implementation time corresponding to the node, and the status information is the data content or intervention content recorded by the node.

[0060] Time and status information are extracted from each node in the intervention feedback chain. For intervention nodes, such as the intervention implementation node, the time information is the implementation time of the intervention, and the status information is the specific content of the intervention, including the intervention type, intensity, and method. For intervention adjustment nodes, the time information is the adjustment time, and the status information is the specific content of the adjustment, such as the magnitude of dose adjustment or changes in training frequency. For intervention termination nodes, the time information is the termination time, and the status information is the reason and method of termination. For data nodes, such as the neuroimaging node, the time information is the image acquisition time, and the status information is the imaging examination results, such as brain region volume measurements and gray matter density values. For cognitive assessment nodes, the time information is the assessment time, and the status information is the scale score data. For follow-up record nodes, the time information is the follow-up time, and the status information is the patient's symptom description and daily living ability records. The extracted time and status information are stored in the node attributes to ensure that each node has complete time and status records.

[0061] Step S132: Sort all nodes in chronological order to form a time-series node sequence, which contains an alternating arrangement of intervention nodes and data nodes.

[0062] The time information of all nodes in the intervention feedback chain is collected, and then these nodes are sorted based on timestamps. During the sorting process, the year of the node time information is compared first, with nodes of younger years listed first; if the years are the same, the month is compared, with younger months listed first; if the months are also the same, the date is compared, with younger dates listed first; if the dates are still the same, the specific time (hour, minute, second) is compared. Through this method, all nodes are arranged into a linear time-series node sequence according to their chronological order. In this sequence, intervention nodes and data nodes alternate. For example, the sequence might start with the pre-intervention node (data node), followed by the intervention type node, the intervention implementation node (intervention node), then the post-intervention node (data node), then the intervention adjustment node (intervention node), then the adjusted post-intervention node (data node), and finally the intervention termination node (intervention node) and the corresponding final state node (data node), etc. This alternating time-series node sequence reflects the temporal relationship between the intervention implementation process and the data change process.

[0063] Step S133: Select adjacent data nodes of different types in the time sequence and analyze the matching degree between the evolution trend of the state information of the preceding data node and the evolution trend of the state information of the following data node.

[0064] Step S1331: Extract the time sequence of nodes formed by sorting them in chronological order, identify all data nodes in the time sequence that are marked as neuroimaging nodes, cognitive assessment nodes and follow-up record nodes, exclude intervention type nodes, intervention implementation nodes, intervention adjustment nodes and intervention termination nodes in the time sequence, and obtain a pure data node sequence.

[0065] In the time-series node sequence, each node has its type identifier. By traversing the entire time-series node sequence, data nodes are filtered out based on the node type identifier, namely neuroimaging nodes, cognitive assessment nodes, and follow-up record nodes, while intervention nodes (intervention type nodes, intervention implementation nodes, intervention adjustment nodes, and intervention termination nodes) are excluded. For example, if the time-series node sequence is [pre-intervention neuroimaging node, intervention type node, intervention implementation node, post-intervention cognitive assessment node, post-intervention follow-up record node, intervention adjustment node, post-intervention neuroimaging node, intervention termination node, final state cognitive assessment node], the pure data node sequence obtained after filtering is [pre-intervention neuroimaging node, post-intervention cognitive assessment node, post-intervention follow-up record node, post-intervention neuroimaging node, final state cognitive assessment node].

[0066] Step S1332: Select adjacent data node combinations from the pure data node sequence. An adjacent data node combination refers to two data nodes that are in consecutive positions in the pure data node sequence and that are of different types. The data node types include neuroimaging node type, cognitive assessment node type, and follow-up record node type. The selection yields a set of adjacent data node pairs of different types.

[0067] The sequence of pure data nodes is traversed, starting from the first data node and comparing it sequentially with each subsequent data node to determine if they are adjacent and of different types. For example, in the sequence [neuroimaging node A, cognitive assessment node B, follow-up record node C, neuroimaging node D, cognitive assessment node E], two consecutive data node combinations are (neuroimaging node A, cognitive assessment node B), (cognitive assessment node B, follow-up record node C), (follow-up record node C, neuroimaging node D), and (neuroimaging node D, cognitive assessment node E). Then, it is determined whether the two data nodes in each combination are of different types. Since all combinations in the above sequence have different node types, they are all selected into the set of adjacent pairs of different types of data nodes. If two adjacent data nodes have the same type, such as (neuroimaging node A, neuroimaging node B), they are excluded from this set.

[0068] Step S1333: For each data node pair in the set of adjacent data node pairs of different types, determine the preceding data node and the following data node. The preceding data node is the node in the data node pair that is earlier in the time sequence of the time nodes, and the following data node is the node in the data node pair that is later in the time sequence of the time nodes.

[0069] For each pair of data nodes in the set of adjacent data node pairs of different types, compare the time information of the two data nodes. The node with earlier time information is determined as the pre-order data node, and the node with later time information is determined as the post-order data node. For example, for the data node pair (neuroimaging node A, cognitive assessment node B), the time of neuroimaging node A is T1, and the time of cognitive assessment node B is T2. If T1 < T2, then neuroimaging node A is the pre-order data node and cognitive assessment node B is the post-order data node.

[0070] Step S1334: Extract the status information of the pre-order data node. The status information of the pre-order data node is the content of the status change of the corresponding data type recorded by the pre-order data node, and arrange it in the chronological order of its own acquisition time to form a pre-order node status information sequence; extract the status information of the post-order data node. The status information of the post-order data node is the content of the status change of the corresponding data type recorded by the post-order data node, and arrange it in the chronological order of its own acquisition time to form a post-order node status information sequence.

[0071] The pre-order data node may contain status information collected multiple times. For example, a neuroimaging node may correspond to multiple magnetic resonance imaging examinations, and each examination has corresponding status information (such as brain region volume values). Arrange the above status information in the chronological order of acquisition time to form a sequence, that is, the pre-order node status information sequence. Similarly, the post-order data node may also contain status information collected multiple times, and arrange it in the chronological order of acquisition time to form a post-order node status information sequence. For example, if the pre-order data node is a neuroimaging node, and the brain region volume values collected at different times are V1, V2, V3 (arranged in the chronological order of acquisition), then the pre-order node status information sequence is [V1, V2, V3]; if the post-order data node is a cognitive assessment node, and the scale scores collected at different times are S1, S2, S3, then the post-order node status information sequence is [S1, S2, S3].

[0072] Step S1335: Perform time alignment processing on the pre-order node status information sequence and the post-order node status information sequence. Based on the global time order of the time series node sequence, determine the corresponding relationship between the time points corresponding to each status information in the pre-order node status information sequence and the time points corresponding to each status information in the post-order node status information sequence, and generate an aligned double-node status sequence pair.

[0073] The global time order of the time sequence of nodes serves as the basis for sorting all nodes, encompassing the time points corresponding to all state information of both preceding and succeeding data nodes. Starting with the first state information time point in the preceding node's state information sequence, the closest state information time point in the succeeding node's state information sequence is found within the global time order, and these are used as a pair for alignment. Then, using the second state information time point in the preceding node's state information sequence as a reference, the next closest time point in the succeeding node's state information sequence is found within the global time order, and so on, until all state information in both the preceding and succeeding node's state information sequences is time-aligned. If the two sequences have different lengths, for the remaining state information in the longer sequence, the closest time point in the global time order is found for alignment, or interpolation is performed as needed (if the state information is numerical data). Through this method, aligned pairs of two-node state sequences are generated, where the two state information points in each pair come from the preceding and succeeding data nodes and have corresponding time points.

[0074] Step S1336: Extract the state information evolution trend of the preceding data node from the aligned two-node state sequence pair. The state information evolution trend is described by the change characteristics of continuous state information in the preceding node state information sequence. The change characteristics include the continuation, transformation, strengthening and weakening of state information, forming the preceding node trend characteristics. Extract the state information evolution trend of the following data node from the aligned two-node state sequence pair. In the same way as extracting the state information evolution trend of the preceding data node, the following node trend characteristics are formed by describing the change characteristics of continuous state information in the following node state information sequence.

[0075] For a sequence of preceding node state information, the changes in adjacent state information are compared sequentially. If the change in the subsequent state information is small and remains within a certain range compared to the preceding state information, the change characteristic is "continuation"; if the subsequent state information undergoes a significant change in nature compared to the preceding state information, such as changing from an upward trend to a downward trend, the change characteristic is "transformation"; if the direction of change in the subsequent state information is consistent with the previous one but the magnitude of change increases, the change characteristic is "reinforcement"; if the direction of change is consistent with the previous one but the magnitude of change decreases, the change characteristic is "weakening". By describing the change characteristics of continuous state information in the entire preceding node state information sequence, the preceding node trend characteristics are formed. For example, if the change characteristics of the preceding node state information sequence are continuation, reinforcement, transformation, and weakening in sequence, then the preceding node trend characteristics are a sequence containing these change characteristics. The same method is used to analyze the subsequent node state information sequence, extract change characteristics, and form the subsequent node trend characteristics.

[0076] Step S1337: Determine the comparison dimensions of trend features. The comparison dimensions include change direction features, change pattern features, and change persistence features. Change direction features reflect the positive, negative, or stable trend of state information evolution. Change pattern features reflect the gradual, abrupt, or intermittent characteristics of state information evolution. Change persistence features reflect the continuous, interrupted, or repetitive characteristics of state information evolution.

[0077] The direction of change refers to the overall trend of state information over a period of time. If the value of state information gradually increases or develops in a positive direction (such as an increase in cognitive assessment scores, a slower rate of brain region volume reduction, etc.), it is considered "positive"; if the value of state information gradually decreases or develops in a negative direction (such as a decrease in cognitive assessment scores, an accelerated rate of brain region volume reduction, etc.), it is considered "negative"; if the value of state information fluctuates within a certain range without a clear overall trend, it is considered "stable". The pattern of change describes the way state information changes. "Gradual" means that the change in state information is slow and continuous, with a relatively uniform amplitude; "abrupt" means that the state information undergoes a large change in a short period of time; "intermittent" means that the change in state information exhibits characteristics of alternating change and stagnation. The duration of change describes the continuity of state information change. "Continuous" means that the change in state information occurs uninterruptedly over a period of time; "interrupted" means that the change in state information stops during the process and then starts again; "repetitive" means that the direction of change in state information reverses multiple times, such as rising then falling, then rising again and falling again, etc.

[0078] Step S1338: Compare the trend features of preceding nodes with those of subsequent nodes in terms of the direction of change feature dimension, check whether the direction of change features of preceding nodes and subsequent nodes are consistent, and generate a direction feature consistency result; compare the trend features of preceding nodes with those of subsequent nodes in terms of the change pattern feature dimension, check whether the change pattern features of preceding nodes and subsequent nodes are consistent, and generate a pattern feature consistency result; compare the trend features of preceding nodes with those of subsequent nodes in terms of the continuity of change feature dimension, check whether the continuity of change features of preceding nodes and subsequent nodes are consistent, and generate a continuity feature consistency result.

[0079] When comparing the direction of change features, the direction of change at each corresponding time point in the trend features of preceding and subsequent nodes is compared one by one. If the direction of change is the same at most time points (e.g., both are positive, both are negative, or both are stable), the consistency result of the direction feature is "highly consistent"; if they are the same at some time points and different at others, it is "partially consistent"; if they are different at most time points, it is "inconsistent". When comparing the pattern of change features, the change pattern sequences of the trend features of preceding and subsequent nodes are compared. If the change patterns (gradual, abrupt, intermittent) of the two have a high degree of matching at the corresponding time points, the consistency result of the pattern feature is "highly consistent"; if the degree of matching is moderate, it is "partially consistent"; if the degree of matching is low, it is "inconsistent". When comparing the persistence of change features, the persistence of change sequences of the trend features of the preceding and succeeding nodes are compared. If the persistence of change (continuous, interrupted, repeated) of the two is highly consistent at the corresponding time point, the consistency result of the persistence feature is "highly consistent"; if the consistency is moderate, it is "partially consistent"; if the consistency is low, it is "inconsistent".

[0080] Step S1339: Summarize the consistency results of directional features, pattern features, and persistence features to form a summary result of trend feature consistency; determine the matching degree category based on the summary result of trend feature consistency, and check the time difference between the start time and the turning time of the state change in the preceding node state information sequence and the following node state information sequence. If the time difference is within a reasonable range of the collection interval between the preceding and following data nodes, the determined matching degree category is maintained; if the time difference exceeds the reasonable range, the matching degree category level is reduced by one level. The reasonable range is determined based on the regular collection cycle of the corresponding data type; bind the finally determined matching degree category with the corresponding adjacent data node pairs of different types, supplement the type combination information of the data node pair and the aligned double-node state sequence pair information, and generate the trend matching degree analysis result for each data node pair.

[0081] The trend feature consistency summary result comprehensively considers the consistency results of directional features, pattern features, and persistence features. For example, if the consistency results of all three dimensions are "highly consistent," the trend feature consistency summary result is "excellent"; if two dimensions are "highly consistent" and one dimension is "partially consistent," the summary result is "good"; if one dimension is "highly consistent," two dimensions are "partially consistent," or all three dimensions are "partially consistent," the summary result is "moderate"; if there are "inconsistent" dimensions, the summary result is determined as "low" or "poor" based on the number and severity of the "inconsistent" dimensions. Based on the summary result, a preliminary matching category is determined, such as "excellent" corresponding to "highly matched," "good" corresponding to "moderately matched," "moderate" corresponding to "low matched," and "low" and "poor" corresponding to "no match." Then, the time difference between the start time and the turning point time of state changes in the preceding and following node state information sequences is examined. For neuroimaging data, the typical acquisition period may be three to six months, and the reasonable time difference range may be 20% to 30% of the acquisition period. For cognitive assessment data, the typical acquisition period may be one to three months, and the reasonable time difference range may be 15% to 25% of the acquisition period. For follow-up record data, the typical acquisition period may be one week to one month, and the reasonable time difference range may be 10% to 20% of the acquisition period. If the time difference is within a reasonable range, the initially determined matching degree category is maintained; if it exceeds the reasonable range, the matching degree category level is reduced by one level. For example, initially "highly matched," it is reduced to "moderately matched" after exceeding the range. The final matching degree category is bound to the data node pair, and type combination information (such as neuroimaging node-cognitive assessment node) and aligned two-node state sequence pair information are supplemented to form the trend matching degree analysis result.

[0082] Step S134: If the matching degree of the evolution trend of the preceding data node and the following data node exceeds the preset standard, then mark that there is a trend correlation between the two and record it as a potential implicit correlation.

[0083] The preset standards are set based on experience in the field of research on cognitive impairment in the elderly and the needs of data analysis. For example, matching categories of "highly matched" and "moderately matched" are set to exceed the preset standards. For the trend matching results of each pair of adjacent data nodes of different types, if the matching category is "highly matched" or "moderately matched," it is considered that the evolution trend matching degree between the preceding and following data nodes exceeds the preset standards, and a trend correlation is marked between the two. When recording this potential implicit correlation, it is necessary to include the identifier of the preceding data node, the identifier of the following data node, the combination of data node pair types, the trend matching analysis results (matching category, time difference check, etc.), and the aligned two-node state sequence information, so as to verify and analyze the potential implicit correlation later.

[0084] Step S135: Select non-adjacent data nodes in the time-series node sequence that are separated by one or more nodes, and analyze the correlation of their state information over a longer time span, including the correlation of triggering conditions, maintenance conditions and decay conditions of state changes. If there are shared triggering conditions or maintenance conditions, mark the two as having a conditional correlation relationship and record it as a potential latent correlation relationship.

[0085] Step S1351: Determine the non-adjacent data nodes in the time-series node sequence that are separated by one or more nodes, including two neuroimaging nodes of the interval intervention adjustment node, and the neuroimaging nodes and follow-up record nodes of the interval cognitive assessment node.

[0086] In a time-series node sequence, starting from the first data node, other data nodes (which can be intervention nodes or data nodes) are searched sequentially, separated by one or more nodes. For example, in a time-series node sequence [data node A, intervention node 1, data node B, data node C, intervention node 2, data node D], data node A and data node B are separated by intervention node 1, making them non-adjacent data nodes; data node A and data node C are separated by intervention node 1 and data node B, making them non-adjacent data nodes; data node B and data node D are separated by data node C and intervention node 2, and so on. Special attention is paid to two neuroimaging nodes between intervention adjustment nodes, such as [neuroimaging node 1, intervention adjustment node, neuroimaging node 2], where these two neuroimaging nodes are non-adjacent data nodes; and to neuroimaging nodes between cognitive assessment nodes and follow-up record nodes, such as [neuroimaging node 3, cognitive assessment node, follow-up record node 1], where neuroimaging node 3 and follow-up record node 1 are non-adjacent data nodes.

[0087] Step S1352: Extract the triggering conditions for the state changes of preceding non-adjacent data nodes. The triggering conditions include three categories: the implementation of specific intervention measures, changes in specific physiological indicators, and changes in specific lifestyle behaviors. Extract the triggering conditions for the state changes of subsequent non-adjacent data nodes using the same extraction method as for preceding non-adjacent data nodes.

[0088] For preceding non-adjacent data nodes, analyze the relevant information before and after the point in time when their state information begins to change. If a new intervention was implemented before that point in time (e.g., starting medication, cognitive training), then "implementation of a specific intervention" is a trigger condition; if a patient's physiological indicator (e.g., blood pressure, blood sugar, blood lipids) changed significantly before or after that point in time, then "change in a specific physiological indicator" is a trigger condition; if the patient's lifestyle behaviors (e.g., changes in eating habits, increased exercise frequency, improved sleep quality) changed, then "change in a specific lifestyle behavior" is a trigger condition. The same method is used to extract trigger conditions for state changes in subsequent non-adjacent data nodes.

[0089] Step S1353: Compare the triggering conditions of the preceding non-adjacent data nodes and the following non-adjacent data nodes. If there are completely identical or essentially identical triggering conditions, it is determined that the preceding non-adjacent data nodes and the following non-adjacent data nodes have a triggering condition association relationship.

[0090] The trigger condition sets for preceding non-adjacent data nodes are compared with those for subsequent non-adjacent data nodes. If both sets contain identical trigger conditions, such as both including the intervention trigger condition "start using drug A," a trigger condition association is determined. If the trigger conditions are described differently but are essentially the same, for example, the trigger condition for the preceding node is "increased daily exercise time," and the trigger condition for the subsequent node is "increased weekly exercise frequency," both of which belong to exercise-related changes in lifestyle behavior and are essentially the same, then a trigger condition association is also determined.

[0091] Step S1354: Extract the maintenance conditions required for the state changes of preceding non-adjacent data nodes. The maintenance conditions include three categories: continuous implementation of intervention measures, stability of environmental factors, and cooperation of accompanying treatments. Extract the conditions required for the state changes of subsequent non-adjacent data nodes using the same extraction method as for preceding non-adjacent data nodes.

[0092] After a state change in a preceding non-adjacent data node begins, certain conditions are required to maintain that change. If the state change is caused by an intervention, then "continuous implementation of the intervention" is one of the maintenance conditions. If the patient's living environment (such as housing, nursing staff, etc.) remains stable, then "stable environmental factors" is one of the maintenance conditions. If the patient is simultaneously receiving other concomitant treatments (such as treatment for comorbidities like hypertension and diabetes), and the coordination of these concomitant treatments helps maintain the current state change, then "coordination of concomitant treatments" is one of the maintenance conditions. Similarly, the conditions required to maintain the state changes of subsequent non-adjacent data nodes are extracted.

[0093] Step S1355: Compare the maintenance conditions of the preceding non-adjacent data nodes and the following non-adjacent data nodes. If there are completely identical or essentially identical maintenance conditions, it is determined that the preceding non-adjacent data nodes and the following non-adjacent data nodes have a maintenance condition association relationship.

[0094] Compare the maintenance condition sets of preceding and subsequent non-adjacent data nodes. If there are identical maintenance conditions, such as both requiring "continuous use of drug B" (continuous implementation of the intervention), then a maintenance condition association is determined to exist. If the maintenance conditions are essentially the same, for example, the maintenance condition of the preceding node is "fixed home caregiver" and the maintenance condition of the subsequent node is "stable nursing team," both belonging to the stability of environmental factors, then a maintenance condition association is also determined to exist.

[0095] Step S1356: Extract the fading conditions for the fading of state changes in preceding non-adjacent data nodes. The fading conditions include three categories: termination of intervention measures, occurrence of interfering factors, and deterioration of physical condition. Extract the fading conditions for the fading of state changes in subsequent non-adjacent data nodes using the same extraction method as for preceding non-adjacent data nodes.

[0096] When the state change of a non-adjacent data node stops or reverses, analyze the reasons for its decay. If the state change decays due to the cessation of intervention, then "termination of intervention" is the decay condition; if the state change decays due to the appearance of new interfering factors (such as infection, trauma, or mental stress), then "appearance of interfering factors" is the decay condition; if the patient's overall physical condition deteriorates (such as the appearance of a new serious illness or organ failure), then "deterioration of physical condition" is the decay condition. The same method is used to extract the decay conditions for the state changes of subsequent non-adjacent data nodes.

[0097] Step S1357: If there is any one of the following conditions: triggering condition association, maintaining condition association, or fading condition association between the preceding non-adjacent data node and the following non-adjacent data node, then mark the preceding non-adjacent data node and the following non-adjacent data node as having a condition association relationship and record it as a potential implicit association relationship.

[0098] If a preceding and succeeding non-adjacent data node is associated with any one of the triggering, maintenance, or fading conditions, then a conditional association is determined to exist between them. When recording this potential implicit association, the identifiers of the preceding and succeeding non-adjacent data nodes, the type of association (triggering condition association, maintenance condition association, or fading condition association), the specific content of the association conditions, and relevant information about state changes are included for subsequent verification.

[0099] Step S136: Select different types of data nodes in the time-series node sequence that are associated with the same intervention node, analyze the order of state changes and mutual influence among the different types of data nodes associated with the same intervention node, and if the state change of any data node precedes the changes of other data nodes and has a predictive effect on the changes of other data nodes, then mark that there is a predictive relationship between any data node and other data nodes, and record it as a potential latent relationship.

[0100] First, identify each intervention node in the time-series node sequence. Then, identify all data nodes associated with that intervention node. These data nodes are typically post-intervention nodes and are of different types (neuroimaging nodes, cognitive assessment nodes, and follow-up record nodes). For example, an intervention implementation node may be associated with neuroimaging node A, cognitive assessment node B, and follow-up record node C. Analyze the temporal order of state changes in these different types of data nodes and compare their start times. If the start time of state change in neuroimaging node A is earlier than that of cognitive assessment node B and follow-up record node C, and analysis of historical data reveals that when neuroimaging node A experiences a specific state change, cognitive assessment node B and follow-up record node C often experience corresponding state changes in the following period, then the state change of neuroimaging node A is considered to have a predictive effect on the changes in cognitive assessment node B and follow-up record node C. In this case, mark the predictive association between neuroimaging node A and cognitive assessment node B, and between neuroimaging node A and follow-up record node C. When recording this potential implicit relationship, it includes the identifier of the predictive data node, the identifier of the predicted data node, the sequence of state changes, the specific manifestation of the predictive effect, and the relevant intervention node information.

[0101] Step S137: Verify the potential implicit relationships in all records. The verification process includes cross-validating the occurrence of the same type of potential implicit relationships in the intervention feedback chain of different research subjects, and verifying whether there are data support vulnerabilities in the node status information corresponding to the potential implicit relationships, and obtaining the verification results.

[0102] The intervention feedback loops of multiple research subjects were collected. For each type of potential implicit association (trend association, conditional association, predictive association), its frequency of occurrence in the intervention feedback loops of different research subjects was statistically analyzed. If a certain potential implicit association appeared in the intervention feedback loops of multiple research subjects, the cross-validation result was "widely present"; if it appeared only in some research subjects, it was "partially present"; and if it appeared only in a few research subjects, it was "individually present". Simultaneously, the completeness and accuracy of the node state information corresponding to the potential implicit associations were checked, and whether there were any contradictory or missing data. For example, if a potential implicit association describes the state change of a data node, but the state information of that data node is incomplete or contains obvious errors, then there is a data support gap. Based on the cross-validation results and the data support gap check results, the validation results were obtained, including categories such as "validation passed", "validation partially passed", and "validation failed". "Verification passed" indicates that the cross-validation result is "widely present" and there is no data to support the vulnerability; "Verification partially passed" indicates that the cross-validation result is "partially present" or there is a minor data to support the vulnerability; "Verification failed" indicates that the cross-validation result is "individually present" or there is a serious data to support the vulnerability.

[0103] Step S138: Based on the verification results, eliminate potential latent associations that fail verification, retain potential latent associations that pass verification, and classify them into three categories according to association type: trend association, conditional association, and predictive association, to form a set of latent temporal associations.

[0104] For potential implicit associations whose verification result is "verification failed," they are removed from the list. For potential implicit associations whose verification result is "verification passed," they are classified according to their association type (trend association, conditional association, predictive association). During the classification process, a subset is created for each type of association, and each subset contains all verified potential implicit associations of that type. Each potential implicit association in the subset contains complete node identifiers, association conditions, status information, time information, etc. Through the above classification and organization, a well-structured set of implicit temporal associations is formed. This set of implicit temporal associations contains different types of potential implicit associations, reflecting the potential associations between different data nodes that evolve over time.

[0105] Step S140: Based on the implicit temporal correlation set, perform hierarchical aggregation of the intervention feedback links of multiple research subjects, extract the common patterns applicable to the group and the individual difference patterns that exist only in a single research subject, and form a set of group individual patterns.

[0106] Step S141: The intervention feedback links of multiple research subjects are initially classified according to the intervention type to form different intervention type groups. Each intervention type group contains the intervention feedback links of research subjects who adopted the same intervention measures.

[0107] All intervention feedback links for research subjects were collected. Each intervention feedback link contained an intervention type node, which recorded the type of intervention. Based on the content of the intervention type node, intervention feedback links of research subjects with the same intervention type were grouped together. For example, intervention feedback links of all research subjects using drug treatment (such as donepezil) as the primary intervention were grouped together as the drug treatment group; intervention feedback links of research subjects using cognitive training as the primary intervention were grouped together as the cognitive training group; and those using rehabilitation therapy were grouped together as the rehabilitation therapy group, and so on. For research subjects using multiple interventions simultaneously, the type of their primary intervention was used as the classification basis. If the primary intervention could not be clearly identified, it was determined based on the duration or intensity of the intervention, or they were separately listed as a comprehensive intervention group. Through the above classification method, multiple different intervention type groups were formed.

[0108] Step S142: For each intervention type group, match the intervention feedback links of all research subjects in the group with the set of implicit time-series associations, and calculate the number of times each implicit time-series association appears in the intervention feedback links within the group.

[0109] For each intervention type group, the intervention feedback loop for each subject within the group is traversed. For each intervention feedback loop, the existence of various potential latent associations from the latent temporal association set is examined. During the examination, the node pairs in the intervention feedback loop are compared with the node pairs in the latent temporal association set to see if they share the same type combination, association conditions, and state change characteristics. If a matching latent latent association exists, its occurrence in the intervention feedback loop is recorded. After traversing all intervention feedback loops for all subjects within the group, the total number of occurrences of each latent temporal association in the group is counted. For example, in the drug treatment group, the number of occurrences of trend association A, conditional association B, and predictive association C are counted.

[0110] Step S143: Based on the proportion of occurrences to the total number of subjects in the group, screen out the implicit temporal associations whose proportions exceed a preset threshold, integrate the implicit temporal associations whose proportions exceed the preset threshold, and form a set of common associations for the intervention type group.

[0111] The preset threshold is determined based on research needs and the size of the intervention group; for example, the preset threshold is 50%. For each implicit temporal association, the ratio of its occurrence frequency to the total number of subjects in the group (occurrence percentage) is calculated. If this ratio exceeds the preset threshold, the implicit temporal association is considered to be prevalent in the intervention group and is selected into the group common association set. For example, if the drug treatment group has 100 subjects and a certain implicit temporal association occurs 60 times, the occurrence percentage is 60%, exceeding the preset threshold of 50%, then it is integrated into the group common association set of the drug treatment group. All implicit temporal associations that meet the criteria are integrated together to form the group common association set for each intervention group.

[0112] Step S144: Analyze the temporal distribution characteristics of each latent temporal association in the common association set of the group, determine the typical occurrence stages of different latent temporal associations within the intervention period, extract the core association relationships of each typical occurrence stage, and form the common patterns of the group.

[0113] Step S1441: Extract the time information corresponding to each implicit temporal association in the set of common associations of the group. The time information comes from the time records of the intervention nodes and data nodes involved in the implicit temporal association. The time records of the intervention nodes include the implementation time of the intervention implementation node and the adjustment time of the intervention adjustment node. The time records of the data nodes include the acquisition time of the neuroimaging node, the assessment time of the cognitive assessment node, and the tracking time of the follow-up record node.

[0114] For each implicit temporal association in the set of common associations in the group, find all intervention nodes and data nodes involved. Extract the implementation time of the intervention implementation node and the adjustment time of the intervention adjustment node from the intervention nodes. Extract the acquisition time of the neuroimaging node, the assessment time of the cognitive assessment node, and the tracking time of the follow-up record node from the data nodes. Organize the above time records into a time information list, where each time information includes a timestamp and the corresponding node type. For example, if an implicit temporal association involves the intervention implementation node (implementation time T1), the neuroimaging node (acquisition time T2), and the cognitive assessment node (assessment time T3), then its time information list is [T1 (intervention implementation), T2 (neuroimaging acquisition), T3 (cognitive assessment)].

[0115] Step S1442: Match the time information associated with each implicit time series with the corresponding intervention cycle. The intervention cycle is defined by the implementation time of the intervention implementation node and the termination time of the intervention termination node. The time span of the intervention cycle is formed by taking the implementation time of the intervention implementation node as the starting point and the termination time of the intervention termination node as the ending point.

[0116] For each implicit time series association, its corresponding intervention period is determined. The intervention period refers to the entire time span from the implementation of the intervention to its termination, starting at the implementation time of the intervention implementation node and ending at the termination time of the intervention termination node. Each time point in the implicit time series association's time information list is compared with the time span of the intervention period to determine if that time point is within the intervention period. For example, if the intervention implementation time is Tstart and the intervention termination time is Tend, the resulting intervention period's time span is [Tstart, Tend]. If a certain time information T in the implicit time series association falls within this span (Tstart ≤ T ≤ Tend), then that time information is considered to match the intervention period.

[0117] Step S1443: Sort the time information of all latent time series associations according to the time span of the intervention period, and generate the time distribution sequence of each latent time series association within the intervention period. The time distribution sequence includes the occurrence time of the latent time series association and the association type identifier. The association type identifier corresponds to trend association, conditional association, and predictive association.

[0118] Based on the time span of the intervention period, the time information of each implicit time series association is sorted in chronological order. The sorted time information constitutes a time distribution sequence, in which each element contains the occurrence time of the implicit time series association (i.e., the timestamp in the time information) and an association type identifier (trend association, conditional association, or predictive association). For example, the time distribution sequence of an implicit time series association may be [(T1, trend association), (T2, conditional association), (T3, predictive association)], where T1, T2, and T3 are the timestamps arranged in chronological order within the intervention period.

[0119] Step S1444: Extract the time stamps of intervention nodes within the intervention cycle. Using the time of the intervention implementation node, intervention adjustment node, and intervention termination node as the dividing points, divide the intervention cycle into multiple consecutive time periods. Each time period includes a start time and an end time, and the start time and end time correspond to the time of two adjacent intervention nodes, respectively.

[0120] The intervention node time marks within the intervention period include the implementation time of the intervention implementation node, the adjustment times of all intervention adjustment nodes, and the termination time of the intervention termination node. Arrange the above time marks in chronological order, such as [T implementation, T adjustment 1, T adjustment 2,..., T termination]. Then, taking two adjacent time marks as demarcation points, divide the intervention period into multiple time periods. The starting time of the first time period is T implementation, and the ending time is T adjustment 1; the starting time of the second time period is T adjustment 1, and the ending time is T adjustment 2; and so on. The starting time of the last time period is the adjustment time of the last intervention adjustment node, and the ending time is T termination. If there are no intervention adjustment nodes within the intervention period, the entire intervention period is one time period, with the starting time being T implementation and the ending time being T termination.

[0121] Step S1445: Classify the implicit time series associations in the time distribution sequence into the corresponding time periods according to the occurrence time, count the number of implicit time series associations corresponding to different association type identifiers in each time period, and generate a time period association distribution result. The time period association distribution result includes the corresponding relationships between each time period and the corresponding implicit time series associations.

[0122] For the occurrence time of each implicit time series association in the time distribution sequence, determine which time period it belongs to. For example, if the occurrence time of a certain implicit time series association is T, if T implementation ≤ T < T adjustment 1, then classify it into the first time period; if T adjustment 1 ≤ T < T adjustment 2, then classify it into the second time period, and so on. Then, count the number of implicit time series associations of each association type identifier (trend association relationship, conditional association relationship, predictive association relationship) by time period. For example, if the trend association relationship appears A times, the conditional association relationship appears B times, and the predictive association relationship appears C times in the first time period, then the association distribution of the first time period in the time period association distribution result is {trend association relationship: A, conditional association relationship: B, predictive association relationship: C}.汇总所有时段的关联分布情况,生成时段关联分布结果。汇总所有时段的关联分布情况,生成时段关联分布结果。Summarize the association distribution situations of all time periods to generate a time period association distribution result.

[0123] Step S1446: Determine the typical occurrence stage according to the time period association distribution result. Select the time period with the highest proportion of the number of implicit time series associations as the typical occurrence stage corresponding to this association type identifier. If the proportion of the number in multiple time periods of the same association type identifier is the same, then determine it in combination with the node types involved in the implicit time series associations corresponding to this association type identifier.优先选取与干预实施节点或干预调整节点时间距离最近的时段。优先选取与干预实施节点或干预调整节点时间距离最近的时段。Preferentially select the time period with the shortest time distance from the intervention implementation node or the intervention adjustment node.

[0124] For each type of association identifier, calculate its proportion in each time period (the number of that association type in a time period divided by the total number of all association types in that time period). Select the time period with the highest proportion as the typical occurrence stage of that association type identifier. For example, if the proportion of trend associations is 60% in time period one, 30% in time period two, and 10% in time period three, then time period one is the typical occurrence stage of trend associations. If the same association type identifier has the same proportion in two or more time periods, for example, 45% in both time period two and time period three, then analyze the node types involved in the implicit time-series associations corresponding to that association type identifier. If the node types involved are strongly related to the intervention implementation node or intervention adjustment node (e.g., the change in the state of data nodes is directly related to the implementation or adjustment of intervention measures), then prioritize selecting the time period closest in time to the intervention implementation node or intervention adjustment node as the typical occurrence stage.

[0125] Step S1447: For each typical occurrence stage, extract all implicit temporal correlations within that stage, and analyze the correlation attributes between each implicit temporal correlation and the intervention node. The correlation attributes of the intervention node include the expected impact direction description of the intervention type node, the expected correlation description of the intervention implementation node, and the correction direction description of the intervention adjustment node.

[0126] For each typical occurrence stage, all implicit temporal correlations within that stage are collected. For each implicit temporal correlation, the intervention nodes involved are identified, and their correlation attributes are obtained. For example, if the implicit temporal correlation involves intervention type nodes, their expected impact direction is described (promoting positive change, inhibiting negative change, etc.); if it involves intervention implementation nodes, their expected correlation description is obtained (increased intensity may enhance the magnitude of change, etc.); if it involves intervention adjustment nodes, their correction direction description is obtained (strengthening positive change, mitigating negative change, etc.). The relationship between these correlation attributes and implicit temporal correlations is analyzed to understand how the correlation attributes of intervention nodes affect the formation and manifestation of implicit temporal correlations.

[0127] Step S1448: Based on the association attributes, select the implicit time-series associations with the highest matching degree with the association attributes of the intervention node. The matching degree is determined by the correspondence between the association type of the implicit time-series association and the association attributes of the intervention node. Trend associations prioritize matching the expected impact direction description, conditional associations prioritize matching the expected association description, and predictive associations prioritize matching the correction direction description. Integrate the selected implicit time-series associations according to the association type and extract the recurring association features in each association type. The association features include the data node types involved in the association, the temporal sequence of the association, and the node state change features corresponding to the association. Bind the integrated association features to the typical occurrence stage to form the core association for each typical occurrence stage. The core association contains the association type, association features, and the corresponding intervention node association attributes.

[0128] For trend-based associations, the degree of matching between the evolution trend of state information and the expected impact direction description of intervention type nodes is compared, and the implicit temporal association with the highest matching degree is selected. For conditional associations, the degree of matching between the conditions of state change and the expected description of intervention implementation node associations is compared, and the association with the highest matching degree is selected. For predictive associations, the degree of matching between the predictive effect and the correction direction description of intervention adjustment node is compared, and the association with the highest matching degree is selected. The selected implicit temporal associations are grouped according to association type, and implicit temporal associations of the same association type are placed together. Each group of implicit temporal associations is analyzed to find the recurring association features. For example, multiple trend-based associations involve neuroimaging nodes and cognitive assessment nodes, and the state change of neuroimaging nodes precedes the state change of cognitive assessment nodes, with the state change direction being positive. The above recurring association features are extracted and bound to the typical occurrence stage to form core associations. For example, the core correlation in a typical stage one might be: a trend correlation involving neuroimaging nodes and cognitive assessment nodes, with the neuroimaging nodes changing first and the cognitive assessment nodes changing later, the state change being positive, and the expected impact direction of the corresponding intervention type nodes being described as promoting positive change.

[0129] Step S1449: Summarize the core relationships of all typical occurrence stages and arrange them in chronological order according to the intervention cycle to form common patterns in the group.

[0130] The core relationships of each typical stage are arranged chronologically according to the intervention cycle, that is, arranged according to the order in which the typical stages appear within the intervention cycle. For example, the core relationships of typical stage one (0-1 months after intervention) are listed first, the core relationships of typical stage two (1-3 months after intervention) are listed in the middle, and the core relationships of typical stage three (3-6 months after intervention) are listed last. Through this arrangement, a complete group common pattern is formed, which reflects the core relationships and their development and changes at different stages throughout the entire intervention cycle.

[0131] For example, step S144a: obtain the common association set of the group and the intervention feedback link of the intervention type group corresponding to the common association set of the group, and extract the intervention implementation node time, intervention adjustment node time and intervention termination node time of each research object from the intervention feedback link to form an intervention time set.

[0132] For each intervention type group corresponding to the common association set of the group, the intervention feedback chain for each research subject within the group is traversed. From each intervention feedback chain, the implementation time of the intervention implementation node, the adjustment time of all intervention adjustment nodes, and the termination time of the intervention termination node are extracted. These time information are then aggregated to form an intervention time set. Each element in the intervention time set contains a research subject identifier and a corresponding intervention time record (implementation time, adjustment time, termination time).

[0133] Step S144b: Using the intervention implementation node time as the base time, convert the time information corresponding to each implicit time series association into an offset time relative to the base time. The offset time is obtained by subtracting the base time from the occurrence time of the implicit time series association, thereby generating a relative time identifier for each implicit time series association.

[0134] For each implicit time series association, the intervention implementation date is used as the baseline date (Tbaseline). Then, each occurrence date (Toccurrence) of this implicit time series association is converted into a relative time identifier, i.e., Toffset = Toccurrence - Tbaseline. For example, if the intervention implementation date is Tbaseline = January 1st of a certain year, and one occurrence date of the implicit time series association is February 15th of a certain year, then Toffset = February 15th - January 1st = 45 days, and the relative time identifier is 45 days. In this way, the time information of all implicit time series associations is uniformly converted into an offset time relative to the intervention implementation date, generating a relative time identifier for comparison and statistics along the time dimension in subsequent analysis.

[0135] Step S144c: Summarize the relative time markers of implicit temporal associations of all research subjects within the same intervention type group, sort them according to the size of the relative time markers, and generate a population relative time series. The population relative time series contains the implicit temporal association number and association type corresponding to each relative time marker.

[0136] The relative time markers of implicit temporal associations among all subjects within the same intervention type group were collected and sorted in ascending order. During the sorting process, each relative time marker was accompanied by a corresponding implicit temporal association number and association type (trend association, conditional association, predictive association). For example, if the relative time markers of implicit temporal associations for multiple subjects were 10 days, 20 days, 15 days, and 25 days, respectively, sorting would yield 10 days, 15 days, 20 days, and 25 days, etc. Each relative time marker corresponds to one or more implicit temporal association numbers and association types, thus generating a population relative time series. This generated population relative time series reflects the distribution of implicit temporal associations within the entire intervention type group along the relative time dimension, based on the intervention implementation time reference.

[0137] Step S144d: Extract the relative time corresponding to the intervention adjustment node time in the intervention time set as a time boundary marker. The time boundary marker divides the relative time series of the group into multiple consecutive time intervals. Each time interval starts with a time boundary marker or reference time and ends with another time boundary marker or termination time.

[0138] Extract the relative times (offset times relative to the intervention implementation time) corresponding to all intervention adjustment node times from the intervention time set. These relative times are the time boundary markers. For example, if the relative times corresponding to the intervention adjustment node times are 30 days, 60 days, etc., then the time boundary markers are 30 days and 60 days. Arrange the baseline time (relative time marker 0 days), time boundary markers (30 days, 60 days, etc.), and the relative time corresponding to the intervention termination node (e.g., 90 days) in the relative time series of the population in sequence to divide the relative time series of the population into multiple consecutive time intervals. For example, if the arranged time points are 0 days (baseline time), 30 days (time boundary marker), 60 days (time boundary marker), and 90 days (termination time), then the time intervals are [0 days, 30 days), [30 days, 60 days), and [60 days, 90 days].

[0139] Step S144e: Count the number of implicit temporal associations in each time interval, calculate the proportion of each type of association in the time interval, and generate interval association statistics. The proportion is obtained by dividing the number of the association type by the total number of implicit temporal associations in the time interval.

[0140] For each defined time interval, iterate through all implicit temporal associations belonging to that interval in the relative time series of the population, and count the number of trend associations, conditional associations, and predictive associations. Then, calculate the proportion of each type of association, i.e., the number of a certain association type divided by the total number of implicit temporal associations in that time interval. For example, in the time interval [0 days, 30 days), the total number of implicit temporal associations is 100, of which the number of trend associations is 40, the number of conditional associations is 30, and the number of predictive associations is 30. Therefore, the proportion of trend associations is 40%, the proportion of conditional associations is 30%, and the proportion of predictive associations is 30%. Organize the above statistical data and proportion results into the interval association statistics.

[0141] Step S144f: Select the association type with the highest proportion in the interval association statistics as the dominant association type for that time interval. If multiple association types have the same proportion and are all the highest, then determine the association type based on the node types involved in the implicit time series associations corresponding to that association type, and prioritize the association type corresponding to the node type that is closest in time to the intervention implementation node or intervention adjustment node.

[0142] In the statistical results of interval correlations, the proportion of the three correlation types is compared, and the correlation type with the highest value is the dominant correlation type for that time interval. For example, if trend correlation accounts for 40%, conditional correlation accounts for 30%, and predictive correlation accounts for 30%, then trend correlation is the dominant correlation type for that time interval. If two or three correlation types have the same proportion and are all the highest, such as trend correlation and conditional correlation both accounting for 35%, then the node types involved in the implicit time-series correlations corresponding to these correlation types are examined. The time distance between the node type and the intervention implementation node or intervention adjustment node is compared, and the correlation type corresponding to the node type with the closest time distance is selected as the dominant correlation type.

[0143] Step S144g: Check whether there is temporal continuity in the implicit temporal associations within each time interval in the interval-association correspondence table. Temporal continuity is determined by whether the interval between the relative time identifiers of adjacent implicit temporal associations is uniform. If the interval is not uniform, the time interval is split into smaller sub-intervals and the statistics and classification are performed again.

[0144] For each time interval in the interval-association table, extract the relative time identifiers of all implicit time series associations and calculate the intervals between adjacent relative time identifiers. If these intervals are roughly equal or within a small fluctuation range, the implicit time series associations within that time interval are considered to have temporal continuity; if the intervals vary greatly, with some intervals being very small and others very large, the temporal continuity is considered poor. For time intervals with poor temporal continuity, based on the distribution of intervals, they are divided into multiple sub-intervals to make the intervals between relative time identifiers of adjacent implicit time series associations within each sub-interval more uniform. For example, if the original time interval is [0 days, 90 days], where the intervals between relative time identifiers of implicit time series associations vary greatly, it can be divided into sub-intervals such as [0 days, 30 days], [31 days, 60 days], and [61 days, 90 days]. Then, for each sub-interval, the number of association types and the dominant association type are re-counted.

[0145] Step S144h: Repeat the splitting and statistical steps until the implicit temporal associations in each sub-interval have temporal continuity. At this point, the sub-interval is the typical occurrence stage within the intervention period. Generate a list of typical occurrence stages, which includes the start relative time, end relative time, dominant association type, and the implicit temporal association to which each typical occurrence stage is classified.

[0146] The temporal continuity of the split sub-intervals is checked again. If discontinuities still exist, the splitting continues until all implicit temporal associations within the sub-intervals show good temporal continuity. After multiple splits and statistical analyses, the resulting sub-intervals represent the typical occurrence stages within the intervention period. The start-relative time, end-relative time, dominant association type of each typical occurrence stage, and all included implicit temporal association information are compiled into a list of typical occurrence stages.

[0147] Step S145: For the intervention feedback link of a single research subject within each intervention type group, extract the implicit temporal associations in the intervention feedback link of the single research subject that are not included in the group common association set. The implicit temporal associations in the intervention feedback link of the single research subject that are not included in the group common association set are individual associations unique to the single research subject.

[0148] Within each intervention type group, the intervention feedback loop for each subject within the group is traversed. The implicit temporal associations in the subject's intervention feedback loop are compared with the set of common associations in the group to identify those implicit temporal associations not included in the set of common associations in the group. These unincluded implicit temporal associations are individual associations unique to the individual subject, reflecting the unique data change associations that distinguish the subject from other subjects during the intervention process.

[0149] Step S146: Analyze the node status information and association attributes corresponding to individual associations, and trace the target factors that lead to individual associations. The target factors include four categories: the basic health status of the research subjects, comorbidities, living environment characteristics, and treatment compliance.

[0150] For the extracted individual associations, a detailed analysis of their corresponding node state information is conducted, including the direction, magnitude, and duration of state changes, as well as the node association attributes, such as descriptions of response timeliness and expected impact direction. Combined with other relevant data of the research subject in the scientific research big data set, the possible factors leading to these individual associations are traced. Basic health status includes the research subject's age, gender, history of underlying diseases (e.g., presence of hypertension, diabetes), and physical function; comorbidities refer to whether the research subject has other diseases concurrently with geriatric cognitive impairment and their severity; living environment characteristics include the research subject's residential environment (e.g., living alone, living with family), community environment, and economic status; treatment compliance includes the research subject's adherence to intervention measures (e.g., whether medication is taken on time, whether training is completed as required), and follow-up participation. Through comprehensive analysis of these factors, the main target factors leading to the individual associations are identified.

[0151] Step S147: Combine individual associations with corresponding target factors to form the individual difference patterns of a single research subject. The individual difference patterns of a single research subject reflect the manifestation of individual associations and the influence of target factors.

[0152] This involves associating specific manifestations of individual associations, such as unique trends in state change or unusual response times, with the traced target factors. It describes how the target factors influence the generation and manifestation of individual associations. For example, a subject's poor baseline health (e.g., multiple underlying diseases) may lead to a slower response time to a certain intervention, thus forming a specific individual association; or a subject's low level of treatment compliance may result in an insignificant intervention effect, generating individual associations that differ from the common patterns of the group. Systematically organizing these associations and modes of influence helps to establish the individual difference patterns for each research subject.

[0153] Step S148: Integrate the common patterns of all intervention type groups and the individual differences of all research subjects, classify and arrange them according to intervention type and association type to form a set of group-individual patterns, and distinguish between content applicable to the group and content specific to the individual in the set of group-individual patterns.

[0154] The study collected common patterns across all intervention types and individual differences among all participants, categorizing them into primary groups based on intervention type (e.g., medication, cognitive training, rehabilitation). Within each intervention type, secondary categories were established based on association type (trend association, conditional association, predictive association). During the categorization process, it was clearly distinguished which content was applicable to the entire group and which was specific to a particular participant. For example, in the common patterns of a certain medication treatment, a positive change in cognitive assessment milestones two months after intervention is applicable to the entire group, while for a participant with severe diabetes, the lack of significant change in cognitive assessment milestones is specific to that individual. Through this categorization and integration, a clear and comprehensive set of group and individual patterns was created.

[0155] Step S150: Based on the set of individual patterns of the group, organize the adaptation relationship between intervention measures and data changes and individual adaptation conditions, and generate a research data analysis report on cognitive impairment in the elderly.

[0156] For example, step S151: Analyze the set of individual patterns in the group, and separate the common patterns of the group from the patterns of individual differences. The common patterns of the group include the typical occurrence stages and core relationships of different intervention type groups, while the patterns of individual differences include the individual relationships of each research subject and the corresponding target factors.

[0157] Data analysis techniques were employed to structure the set of patterns among individuals within a group, separating common group patterns from individual differences. The common group patterns section extracted information on typical stages within the intervention period for different intervention types, such as the start and end times of each stage, the dominant correlation type, and the core relationships at each stage, including trend correlations, conditional correlations, and predictive correlations between intervention measures and data changes. The individual differences section extracted detailed individual correlations for each research subject, such as unique state change characteristics, corresponding target factors, and their influencing mechanisms.

[0158] Step S152: Based on the common patterns of the group, extract the intervention measures content in the intervention type node corresponding to each intervention type group, as well as the data change characteristics of the neuroimaging node, cognitive assessment node, and follow-up record node involved in the core correlation. The data change characteristics include the direction of state change, the triggering conditions of the change, and the order of the change.

[0159] For each intervention type group, the specific details of the intervention measures recorded in the intervention type node are extracted from the common patterns of the group, such as drug name, dosage, and method of use, and cognitive training methods, frequency, and duration. Simultaneously, the direction of state change of the data nodes involved in the core correlations is extracted: whether the change is positive, negative, or stable; whether the triggering condition for the change is the implementation or adjustment of the intervention measures or other external factors; and the temporal order of state changes of different data nodes, for example, whether the neuroimaging node changes first or the cognitive assessment node changes first.

[0160] Step S153: Associate the content of the intervention measures with the corresponding data change characteristics, and determine the fitting relationship description based on the association type of the core association. The fitting relationship description corresponding to the trend association includes the correspondence between the intervention measures and the data change trend. The fitting relationship description corresponding to the condition association includes the correspondence between the intervention measures and the data change conditions. The fitting relationship description corresponding to the predictive association includes the correspondence between the intervention measures and the predictive signals of data change.

[0161] The extracted intervention content is correlated one-to-one with data change characteristics. For trend correlations, the trend of data node state changes after intervention implementation is described as whether it is consistent with or opposite to the expected impact direction of the intervention type node (promoting positive change, inhibiting negative change, etc.), as well as the duration and stability of the above trend. For conditional correlations, the conditions under which data nodes will exhibit corresponding state changes are described (such as the intensity and duration of intervention implementation, specific baseline health status, etc.). For predictive correlations, how data node state changes predict the effect of the intervention or potential adjustment needs, for example, a state change in a follow-up record node predicts a possible positive change in a subsequent cognitive assessment node.

[0162] Step S154: Integrate the fitting relationship descriptions of all intervention type groups, classify them according to the type of intervention, and form a group fitting relationship set. The group fitting relationship set includes the content of the intervention, data change characteristics, fitting relationship description and corresponding typical occurrence stage.

[0163] The descriptions of fit relationships across different intervention type groups are summarized and categorized according to the type of intervention (e.g., drug therapy, cognitive training, rehabilitation therapy, etc.). Under each intervention type, the content of all interventions of that type, corresponding data change characteristics, fit relationship descriptions, and the typical stages in which these fit relationships occur within the intervention period are integrated. For example, under the drug therapy category, interventions involving multiple drugs are included. Each drug has its corresponding cognitive assessment node change characteristics, fit relationship descriptions (e.g., positive trends in trend correlations), and the typical stage after intervention implementation where this fit relationship typically occurs (e.g., short-term response stage, medium-term response stage).

[0164] Step S155: Based on the patterns of individual differences, extract the content of intervention measures and data change characteristics involved in the individual associations of each research subject, as well as the corresponding target factors. The target factors include basic health status, comorbidities, living environment characteristics, and treatment compliance.

[0165] From the patterns of individual differences, specific details of the interventions involved in each study subject's individual associations were extracted, such as adjusted drug dosages and specific training programs, as well as the data change characteristics of corresponding neuroimaging nodes, cognitive assessment nodes, and follow-up record nodes. Simultaneously, specific information on the target factors leading to these individual differences was extracted, such as the specific underlying diseases in the baseline health status, the names and severity of comorbidities, the specific living environment, and the specific manifestations of treatment compliance (e.g., medication adherence rate, training completion rate).

[0166] Step S156: Analyze the relationship between the target factor and the individual, determine the difference between the data change characteristics in the individual relationship and the corresponding data change characteristics in the group fit relationship set, and the specific content of the target factor that caused the difference, and form a description of individual differences.

[0167] Analyzing how target factors influence individual associations reveals several factors. For example, participants with poor baseline health may show smaller-than-average changes in data characteristics when receiving the same intervention; comorbid conditions may delay data changes; environmental factors may enhance or weaken the intervention's effectiveness; and low treatment compliance may result in minimal data changes. Comparing the data change characteristics in individual associations with those in the group fit set for the same intervention type identifies differences, such as variations in the magnitude of change, the timing of change, and trends. The specific target factors causing these differences are then identified. For instance, participant A, due to severe diabetes (target factor), may show a lower increase in cognitive assessment scores (data change characteristics) than the average increase for that intervention in the group fit set. These differences and their causes constitute the description of individual differences.

[0168] Step S157: Based on the description of individual differences, organize the individual fit conditions. The individual fit conditions include the content of applicable intervention measures, the corresponding target factor characteristics, the differences in data change characteristics, and adjustment suggestions. The adjustment suggestions are determined based on the correction direction description of the intervention adjustment node.

[0169] Based on the description of individual differences, determine the individual fit conditions suitable for the research subjects. Applicable interventions may be adjustments to common interventions for the group, such as increasing or decreasing medication dosage or personalizing training programs. The corresponding target factor characteristics are the key factors leading to individual differences, such as specific underlying disease types and severity, or special living environment conditions. Differences in data change characteristics refer to the specific differences in data change between individuals and the group. Adjustment recommendations are proposed based on the description of the correction direction at the intervention adjustment points (e.g., strengthening positive changes, alleviating negative changes). For example, for research subjects whose data changes are not significant due to low treatment compliance, it is recommended to increase the follow-up frequency and provide more supervision and reminders to improve compliance; for research subjects whose changes are delayed due to comorbidities, it is recommended to extend the intervention observation period or adjust the intensity of the intervention.

[0170] Step S158: Classify the individual fit conditions of all research subjects according to the target factor type to form an individual fit condition set. The individual fit condition set includes the target factor characteristics, individual fit conditions, and corresponding individual associations.

[0171] The individual fit criteria for all research subjects were categorized according to the type of target factor (basal health status, comorbidities, living environment characteristics, and treatment compliance). Within each target factor type, further subdivisions were made based on specific factor characteristics (such as different diseases in the basic health status, different types of comorbidities, etc.), compiling the individual fit criteria corresponding to that factor characteristic, including applicable intervention measures, differences in data change characteristics, and adjustment suggestions, while also linking corresponding individual association information. For example, the comorbidity type was divided into subtypes such as comorbid diabetes and comorbid hypertension, with each subtype containing the individual fit criteria and individual associations of research subjects with the corresponding comorbidities.

[0172] Step S159: Establish an association mapping between the set of group fit relationships and the set of individual fit conditions. Each group fit relationship corresponds to multiple individual fit conditions. The association mapping is determined based on the correspondence between the content of the intervention measures and the target factors, and an fit relationship-condition correspondence table is generated.

[0173] By using the content of intervention measures as a connecting link, each group fit relationship in the group fit relationship set is associated with individual fit conditions in the individual fit condition set that have the same intervention content. Simultaneously, the correspondence with target factors is considered; that is, the intervention content upon which the group fit relationship is based will produce different individual fit conditions under the influence of different target factor characteristics. For example, a group fit relationship for a certain drug treatment may correspond to multiple individual fit conditions arising from different comorbidities (target factor characteristics). These associations are presented in tabular form, with each row representing a group fit relationship and each column representing the different individual fit conditions corresponding to that group fit relationship, thus generating a fit relationship-condition correspondence table.

[0174] Step S160: Extract the typical occurrence stages and core relationships in the group fit relationship set, the target factor characteristics and adjustment suggestions in the individual fit condition set, and the correspondence in the fit relationship-condition correspondence table, as the core content of the report.

[0175] The report selects representative stages from the group fit set, such as the short-term, medium-term, and long-term stages after intervention implementation, along with core relationships at each stage, including major data trends and conditional correlations. It also extracts typical characteristics of various target factors from the individual fit condition set, such as common underlying diseases and typical living environment problems, along with corresponding adjustment suggestions. Furthermore, it extracts examples of the correspondence between group fit relationships and individual fit conditions from the fit-condition mapping table. These elements together constitute the core of the report, comprehensively reflecting the group patterns and individual differences in intervention measures and data changes.

[0176] Step S161: Arrange the core content of the report in the order of intervention type, group fit relationship, individual fit condition, and association mapping, and supplement the description of the association attributes of the intervention node corresponding to each intervention. The description of the association attributes of the intervention node includes the description of the expected impact direction, the description of the expected association, and the description of the correction direction.

[0177] When compiling the core content of the report, it is first organized according to different types of interventions, such as drug therapy, cognitive training, and rehabilitation therapy. Under each type of intervention, the group fit relationship is described first, including the overall data change characteristics of this type of intervention, a description of the fit relationship, and typical stages of occurrence. Then, the individual fit conditions are described, categorized by target factor type, introducing different individual fit conditions. Next, the correlation mapping relationship between the group fit relationship and individual fit conditions is shown. Throughout the compilation process, descriptions of the correlation attributes of intervention nodes are added for each intervention, such as the expected direction of impact of intervention type nodes, the expected correlation of intervention implementation nodes, and the correction direction of intervention adjustment nodes, making the report content more complete and detailed.

[0178] Step S162: Perform logical verification on the compiled content to ensure that the correspondence between the intervention measures and the data change characteristics is consistent, that there is no contradiction between the target factors and the individual's fit conditions, and that the time sequence between the typical occurrence stage and the core correlation is coherent.

[0179] During the validation process, it is checked whether the content of each intervention measure is consistent with the data change characteristics of its corresponding group. For example, does a certain cognitive training measure actually lead to a positive change in the cognitive assessment node score? Simultaneously, it is checked whether the causal relationship between the target factor and individual fit conditions is reasonable and whether there are any contradictions, such as the target factor being described as high treatment compliance, but the individual fit conditions showing no significant data change without a reasonable explanation. Furthermore, it is verified whether the time division of typical occurrence stages is accurate and whether the core correlations are consistent in time sequence, such as whether the time of the pre-intervention node is indeed before the intervention implementation, and whether the time of the post-intervention node is after the intervention implementation and matches the time of the intervention adjustment node.

[0180] Step S163: After the verification is passed, generate a research data analysis report on cognitive impairment in the elderly.

[0181] After logical verification and confirmation, the core content of the prepared report should be organized into a standardized report format. The report should include an abstract, introduction, research methods (data collection and integration, intervention feedback loop construction, implicit temporal correlation mining, extraction of group and individual patterns, etc.), results (common patterns of the group, patterns of individual differences, fit relationships and conditions, etc.), discussion (significance of the results, comparison with existing research, limitations, etc.), and conclusions. Ensure the report content is logically clear, the data is accurate, and the expression is standardized.

[0182] Throughout the entire data processing and analysis process, privacy protection principles were strictly adhered to for portions involving privacy-sensitive data. During data collection, all personally identifiable information of research subjects was anonymized, with personally identifiable identifiers removed or replaced. During data storage, encryption technology was used to encrypt and store data, restricting access permissions to only authorized researchers within their authorized scope. Secure transmission protocols were used during data transmission to prevent data leakage. During data analysis and report generation, it was ensured that no personally identifiable information was disclosed; all results were presented in the form of group data or anonymized individual data, thereby effectively protecting the privacy of research subjects and preventing the leakage of privacy-sensitive data.

[0183] Based on the same inventive concept, please refer to Figure 2The diagram shows a schematic block diagram of a big data-based research data analysis system 100 for performing the above-described big data-based research data analysis method for cognitive impairment in the elderly, provided in an embodiment of this application. The big data-based research data analysis system 100 may include a communication unit 110, a machine-readable storage medium 120, and a processor 130.

[0184] In this embodiment, both the machine-readable storage medium 120 and the processor 130 are located in the big data-based research data analysis system for cognitive impairment in the elderly, and are separately configured. Alternatively, the machine-readable storage medium 120 can also be integrated into the processor 130 and can communicate and interact with external systems through the communication unit 110. The machine-readable storage medium 120 stores machine-executable instructions for implementing the scheme of this application, and the processor 130 executes the machine-executable instructions stored in the machine-readable storage medium 120 to implement the big data-based research data analysis method for cognitive impairment in the elderly provided in the aforementioned method embodiments.

[0185] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A data analysis method for research on cognitive impairment in the elderly based on big data, characterized in that, The method includes: Acquire research data on various types of cognitive impairment in the elderly and integrate them to form a research big data set, which includes clinical diagnosis and treatment record data, neuroimaging examination data, cognitive function assessment data, and follow-up tracking data. The intervention-related information and data change information in the aforementioned scientific research big data set are dynamically correlated to construct an intervention feedback link with multiple nodes. Each node in the intervention feedback link corresponds to a type of data unit and the association attribute between the data unit and the intervention measures. The intervention-related information refers to the set of information about the entire process directly related to the diagnosis and treatment intervention behaviors implemented for elderly patients with cognitive impairment, extracted from the clinical diagnosis and treatment record data in the scientific research big data of elderly cognitive impairment. The data change information refers to the set of information extracted from the neuroimaging examination data, cognitive function assessment data, and follow-up tracking data in the scientific research big data of elderly cognitive impairment, used to reflect the evolution of the patient's physiological state, cognitive function, and living status at different time points. The non-explicit correlations between different nodes in the time dimension are mined from the intervention feedback link to form a set of implicit temporal correlations, which contains the potential correlations between different data units that evolve over time. Based on the implicit temporal correlation set, the intervention feedback links of multiple research subjects are hierarchically aggregated to extract common patterns applicable to the group and individual difference patterns that exist only in a single research subject, forming a set of group-individual patterns; Based on the set of patterns of individuals in the group, the adaptation relationship between intervention measures and data changes and individual adaptation conditions are sorted out, and a research data analysis report on cognitive impairment in the elderly is generated. The dynamic correlation between intervention-related information and data change information in the scientific research big data set, and the construction of an intervention feedback chain with multiple nodes, includes: Intervention-related information is extracted from the clinical diagnosis and treatment records of the aforementioned scientific research big data set, and broken down into intervention type nodes, intervention implementation nodes, intervention adjustment nodes, and intervention termination nodes. Each node records the specific content of the corresponding intervention step. Data change information is extracted from the neuroimaging examination data, cognitive function assessment data and follow-up tracking data of the scientific research big data set, and multiple nodes are formed respectively. The multiple nodes include neuroimaging nodes, cognitive assessment nodes and follow-up record nodes. Each node records the state change content of the corresponding data type. Determine the temporal correlation between each intervention node and data node. Based on the time information of the intervention implementation node, forward correlation is established with neuroimaging nodes, cognitive assessment nodes, and follow-up record nodes within a preset time period, which are marked as pre-intervention nodes. For backward correlation, neuroimaging nodes, cognitive assessment nodes, and follow-up record nodes are established from the intervention implementation to the intervention termination period, which are marked as post-intervention nodes. Add association attributes to each node. The association attributes of the pre-intervention node include a description of the degree of influence of the data status of the pre-intervention node on the selection of subsequent intervention measures. The association attributes of the post-intervention node include a description of the response time of the data changes of the post-intervention node and the implementation of intervention measures. The association attributes of the intervention node include a description of the expected direction of the intervention content of the intervention node on the data node. Establish dynamic connections between nodes. Intervention type nodes are connected to intervention implementation nodes and intervention pre-intervention nodes respectively. Intervention implementation nodes are connected to intervention adjustment nodes and intervention post-intervention nodes respectively. Intervention adjustment nodes are connected to intervention post-intervention nodes in subsequent time periods. Intervention termination nodes are connected to the final state node of intervention post-intervention nodes. By leveraging the synergistic effect of node connections and associated attributes, each node in the intervention feedback chain can be traced back to its corresponding intervention node or data node, and each connection is accompanied by a clear description of associated attributes, thereby constructing the intervention feedback chain. The process of mining non-explicit correlations in the time dimension of different nodes from the intervention feedback chain to form a set of implicit time-series correlations includes: Extract the time information and status information of all nodes in the intervention feedback chain. The time information is the collection or implementation time corresponding to the node, and the status information is the data content or intervention content recorded by the node. All nodes are sorted in chronological order to form a time-series node sequence, which contains an alternating arrangement of intervention nodes and data nodes. Select adjacent data nodes of different types in the time-series node sequence and analyze the matching degree between the state information evolution trend of the preceding data node and the state information evolution trend of the following data node. If the matching degree of the evolution trend between the preceding data node and the following data node exceeds the preset standard, then the two are marked as having a trend correlation and recorded as a potential implicit correlation. Select non-adjacent data nodes in the time-series node sequence that are separated by one or more nodes, and analyze the correlation of their state information over a longer time span, including the correlation of triggering conditions, maintenance conditions and decay conditions of state changes. If there are shared triggering conditions or maintenance conditions, mark the two as having a conditional correlation relationship and record it as a potential latent correlation relationship. Select different types of data nodes in the time-series node sequence that are associated with the same intervention node, analyze the order of state changes and mutual influence among the different types of data nodes associated with the same intervention node, and if the state change of any data node precedes the changes of other data nodes and has a predictive effect on the changes of other data nodes, then mark that there is a predictive relationship between any data node and other data nodes, and record it as a potential latent relationship. The potential implicit relationships of all records are verified. The verification process includes cross-validating the occurrence of the same type of potential implicit relationships in the intervention feedback chain of different research subjects, and verifying whether there are data support gaps in the node status information corresponding to the potential implicit relationships, and obtaining the verification results. Based on the verification results, potential latent associations that failed verification were removed, while those that passed verification were retained. These associations were then categorized into three types: trend associations, conditional associations, and predictive associations, forming a set of latent temporal associations.

2. The method for analyzing research data on cognitive impairment in the elderly based on big data according to claim 1, characterized in that, The process involves adding association attributes to each node. The association attributes of the pre-intervention node include a description of the degree of influence of the pre-intervention node's data status on the selection of subsequent intervention measures. The association attributes of the post-intervention node include a description of the response timeliness of the post-intervention node's data changes and the implementation of intervention measures. The association attributes of the intervention node include a description of the expected direction of the intervention content's impact on the data node, including: For the pre-intervention node, analyze the correspondence between the data status of the pre-intervention node and the intervention type in the subsequent intervention type node, calculate the frequency ratio of selecting a specific intervention under the same data status, and determine the degree of influence of the data status on the selection of intervention based on the frequency ratio. The degree of influence includes three categories: dominant choice, auxiliary reference, and no significant influence. For post-intervention nodes, extract the start time of the data change of the post-intervention node and the time information of the corresponding intervention implementation node or intervention adjustment node, calculate the time interval between the two, and determine the response timeliness description based on the time interval. The response timeliness description includes four categories: immediate response, short-term response, medium-term response, and long-term response. Based on research consensus in the field of cognitive impairment in the elderly, the expected impact direction of the intervention type on the data changes of neuroimaging nodes, cognitive assessment nodes, and follow-up record nodes is determined for each intervention type. The expected impact direction is described in four categories: promoting positive changes, inhibiting negative changes, maintaining a stable state, and no clear expectation. For intervention implementation nodes, combined with the expected impact direction of intervention type nodes, supplement the expected correlation between the intensity of intervention implementation and the magnitude of data change. The correlation expectation description includes three categories: increased intensity may increase the magnitude of change, intervention intensity exceeding the threshold may have a reverse effect, and intervention intensity maintained within a specific range can stabilize the effect. For intervention and adjustment nodes, analyze the correlation between the adjustment content and the data changes of subsequent intervention nodes, and determine the correction direction description of the adjustment content on the data changes. The correction direction description includes three categories: strengthening positive changes, mitigating negative changes, and optimizing the stability of changes. For each intervention termination node, the maintenance status of data changes at subsequent intervention nodes after the intervention termination node is calculated to determine the duration of the intervention effect. The duration of the intervention effect includes three categories: rapid decline of effect, short-term maintenance of effect, and long-term maintenance of effect. By binding the above-mentioned association attribute descriptions to the corresponding nodes, the association attributes of each node reflect the association characteristics and influence relationships between the node and other nodes.

3. The method for analyzing research data on cognitive impairment in the elderly based on big data according to claim 2, characterized in that, The step of selecting adjacent data nodes of different types in a time-series node sequence and analyzing the matching degree between the state information evolution trends of preceding data nodes and subsequent data nodes includes: Extract the time sequence of nodes formed by sorting them in chronological order, identify all data nodes in the time sequence that are labeled as neuroimaging nodes, cognitive assessment nodes and follow-up record nodes, and exclude intervention type nodes, intervention implementation nodes, intervention adjustment nodes and intervention termination nodes in the time sequence to obtain a pure data node sequence. Adjacent data node combinations are selected from the pure data node sequence. Adjacent data node combinations refer to two data nodes that are consecutive in the pure data node sequence and have different types. The types of data nodes include neuroimaging node types, cognitive assessment node types, and follow-up record node types. The selection yields a set of adjacent data node pairs of different types. For each data node pair in the set of adjacent data node pairs of different types, determine the preceding data node and the following data node. The preceding data node is the node in the data node pair that is earlier in the time sequence of the time nodes, and the following data node is the node in the data node pair that is later in the time sequence of the time nodes. Extract the status information of the preceding data nodes. The status information of the preceding data nodes is the status change content of the corresponding data type recorded by the preceding data node. Arrange them in the order of the collection time of the preceding data nodes to form a sequence of preceding node status information. Extract the status information of subsequent data nodes. The status information of subsequent data nodes is the status change content of the corresponding data type recorded by the subsequent data node. The subsequent data nodes are arranged in chronological order of their own collection time to form a sequence of subsequent node status information. The preceding node state information sequence and the following node state information sequence are time-aligned. Based on the global time order of the time sequence of the nodes, the correspondence between the time point corresponding to each state information in the preceding node state information sequence and the time point corresponding to each state information in the following node state information sequence is determined, and the aligned two-node state sequence pair is generated. The evolution trend of the state information of the preceding data node is extracted from the aligned two-node state sequence pair. The evolution trend of the state information is described by the change characteristics of the continuous state information in the preceding node state information sequence. The change characteristics include the continuation, transformation, strengthening and weakening of the state information, forming the trend characteristics of the preceding node. Extract the evolution trend of state information of subsequent data nodes from the aligned two-node state sequence pairs. Using the same method as extracting the evolution trend of state information of preceding data nodes, the trend features of subsequent nodes are formed by describing the change characteristics of continuous state information in the state information sequence of subsequent nodes. The comparison dimensions for determining trend characteristics include change direction characteristics, change pattern characteristics, and change persistence characteristics. Change direction characteristics reflect the positive, negative, or stable trend of state information evolution. Change pattern characteristics reflect the gradual, abrupt, or intermittent characteristics of state information evolution. Change persistence characteristics reflect the continuous, interrupted, or repetitive characteristics of state information evolution. The trend features of the preceding node and the trend features of the following node are compared in terms of the direction of change. The direction of change of the trend features of the preceding node and the trend features of the following node are checked to see if they are consistent, and a result of consistency of direction features is generated. The trend features of the preceding nodes and the trend features of the following nodes are compared in terms of the change pattern features. The results are checked to see if the change pattern features of the trend features of the preceding nodes and the change pattern features of the trend features of the following nodes are consistent, and a consistency result of the pattern features is generated. The trend features of the preceding node and the trend features of the following node are compared in terms of the dimension of the change persistence feature. The results are checked to see if the change persistence features of the trend features of the preceding node and the change persistence features of the trend features of the following node are consistent, and a consistency result of persistence features is generated. The consistency results of directional features, pattern features, and persistence features are summarized to form a summary result of trend feature consistency. The matching degree category is determined based on the trend feature consistency summary results. The time difference between the start time and the turning time of the state change in the preceding node state information sequence and the following node state information sequence is checked. If the time difference is within a reasonable range of the collection interval between the preceding data node and the following data node, the determined matching degree category is maintained. If the time difference exceeds the reasonable range, the matching degree category level is reduced by one. The reasonable range is determined based on the regular collection cycle of the corresponding data type. The final determined matching degree category is bound to the corresponding adjacent data node pairs of different types. The type combination information of the data node pair and the aligned double node state sequence information are supplemented to generate the trend matching degree analysis results for each data node pair.

4. The method for analyzing research data on cognitive impairment in the elderly based on big data according to claim 3, characterized in that, The process involves selecting non-adjacent data nodes in a time-series node sequence, spaced one or more nodes apart, and analyzing the correlation of their state information over a longer time span. This includes the correlation of triggering conditions, maintenance conditions, and decay conditions for state changes. If shared triggering conditions or maintenance conditions exist, the two nodes are marked as having a conditional correlation and recorded as a potential latent correlation, including: Identify non-adjacent data nodes in a time-series node sequence that are separated by one or more nodes, including two neuroimaging nodes of an interval intervention adjustment node and two neuroimaging nodes of an interval cognitive assessment node and a follow-up record node. Extract the triggering conditions for changes in the state of preceding non-adjacent data nodes. The triggering conditions include three categories: implementation of specific intervention measures, changes in specific physiological indicators, and changes in specific lifestyle behaviors. Extract the triggering conditions for state changes of subsequent non-adjacent data nodes, using the same extraction method as for preceding non-adjacent data nodes; Compare the triggering conditions of the preceding non-adjacent data nodes and the following non-adjacent data nodes. If there are completely identical or essentially the same triggering conditions, it is determined that the preceding non-adjacent data nodes and the following non-adjacent data nodes have a triggering condition association relationship. Extract the maintenance conditions required to maintain the state changes of preceding non-adjacent data nodes. The maintenance conditions include three categories: continuous implementation of intervention measures, stability of environmental factors, and cooperation of accompanying treatments. Extract the conditions required for the state changes of subsequent non-adjacent data nodes, using the same extraction method as for preceding non-adjacent data nodes; Compare the maintenance conditions of the preceding non-adjacent data nodes and the following non-adjacent data nodes. If there are completely identical or essentially identical maintenance conditions, it is determined that the preceding non-adjacent data nodes and the following non-adjacent data nodes have a maintenance condition association relationship. Extract the fading conditions for the fading of state changes in preceding non-adjacent data nodes. The fading conditions include three categories: termination of intervention measures, appearance of interfering factors, and deterioration of physical condition. Extract the fading conditions of the state changes of subsequent non-adjacent data nodes, using the same extraction method as for preceding non-adjacent data nodes; Compare the fading conditions of the preceding non-adjacent data nodes and the following non-adjacent data nodes. If there are completely identical or essentially the same fading conditions, it is determined that the preceding non-adjacent data nodes and the following non-adjacent data nodes have a fading condition correlation relationship. If a preceding non-adjacent data node and a subsequent non-adjacent data node have any of the following conditions: triggering condition association, maintaining condition association, or fading condition association, then the preceding non-adjacent data node and the subsequent non-adjacent data node are marked as having a condition association relationship, and recorded as a potential implicit association relationship.

5. The method for analyzing research data on cognitive impairment in the elderly based on big data according to claim 1, characterized in that, The method involves hierarchically aggregating the intervention feedback links of multiple research subjects based on the implicit temporal correlation set, extracting common patterns applicable to the group and individual difference patterns existing only in a single research subject, forming a group-specific pattern set, including: The intervention feedback links of multiple research subjects are initially classified according to the intervention type to form different intervention type groups. Each intervention type group contains the intervention feedback links of research subjects who adopted the same intervention measures. For each intervention type group, the intervention feedback links of all research subjects in the group are matched with the set of implicit time-series associations, and the number of occurrences of each implicit time-series association in the intervention feedback links within the group is calculated. Based on the proportion of occurrences to the total number of subjects in the group, implicit temporal associations with a proportion exceeding a preset threshold are selected. These implicit temporal associations with a proportion exceeding the preset threshold are then integrated to form a set of common associations for the intervention type group. Analyze the temporal distribution characteristics of each latent temporal association in the common association set of the group, determine the typical occurrence stages of different latent temporal associations within the intervention period, extract the core association relationships of each typical occurrence stage, and form common patterns of the group; For each intervention type group, for the intervention feedback link of an individual research subject, the implicit temporal associations in the intervention feedback link of that individual research subject that are not included in the group common association set are extracted. The implicit temporal associations in the intervention feedback link of that individual research subject that are not included in the group common association set are individual associations unique to that individual research subject. Analyze the node status information and association attributes corresponding to individual associations, and trace the target factors that lead to individual associations. The target factors include four categories: the basic health status of the research subjects, comorbidities, living environment characteristics, and treatment compliance. By combining individual relationships with corresponding target factors, the individual difference patterns of a single research subject are formed. These individual difference patterns of a single research subject reflect the manifestation of individual relationships and the influence of target factors. The common patterns of all intervention types and the individual differences of all research subjects are integrated and classified according to intervention type and association type to form a set of group-individual patterns. The set of group-individual patterns distinguishes between content applicable to the group and content specific to the individual.

6. The method for analyzing research data on cognitive impairment in the elderly based on big data according to claim 5, characterized in that, The analysis examines the temporal distribution characteristics of each latent temporal association in the common association set of the group, identifies the typical occurrence stages of different latent temporal associations within the intervention period, extracts the core association relationships of each typical occurrence stage, and forms common patterns in the group, including: Extract the time information corresponding to each implicit temporal association in the set of common associations of the group. The time information comes from the time records of the intervention nodes and data nodes involved in the implicit temporal association. The time records of the intervention nodes include the implementation time of the intervention implementation node and the adjustment time of the intervention adjustment node. The time records of the data nodes include the acquisition time of the neuroimaging node, the assessment time of the cognitive assessment node, and the tracking time of the follow-up record node. The time information associated with each implicit time series is matched with the corresponding intervention cycle. The intervention cycle is defined by the implementation time of the intervention implementation node and the termination time of the intervention termination node. The time span of the intervention cycle is formed by taking the implementation time of the intervention implementation node as the starting point and the termination time of the intervention termination node as the ending point. The time information of all latent time series associations is sorted according to the time span of the intervention period to generate a time distribution sequence of each latent time series association within the intervention period. The time distribution sequence includes the occurrence time of the latent time series association and the association type identifier. The association type identifier corresponds to trend association, conditional association, and predictive association. Extract the time stamps of intervention nodes within the intervention cycle. Using the time of intervention implementation node, intervention adjustment node, and intervention termination node as the dividing points, the intervention cycle is divided into multiple consecutive time periods. Each time period includes a start time and an end time, and the start time and end time correspond to the time of two adjacent intervention nodes, respectively. Implicit temporal associations in the time distribution sequence are assigned to corresponding time periods according to their occurrence time. The number of implicit temporal associations corresponding to different association types in each time period is counted, and the time period association distribution results are generated. The time period association distribution results contain the correspondence between each time period and the corresponding implicit temporal association. Based on the distribution results of time period associations, the typical occurrence stage is determined. The time period with the highest proportion of implicit time series associations is selected as the typical occurrence stage corresponding to the association type identifier. If the proportion of the same association type identifier is the same in multiple time periods, it is determined by combining the node types involved in the implicit time series associations corresponding to the association type identifier. Priority is given to selecting the time period that is closest to the intervention implementation node or intervention adjustment node in time. For each typical occurrence stage, all implicit temporal correlations within that stage are extracted, and the correlation attributes of each implicit temporal correlation with the intervention node are analyzed. The correlation attributes of the intervention node include the expected impact direction description of the intervention type node, the expected correlation description of the intervention implementation node, and the correction direction description of the intervention adjustment node. The implicit time-series associations with the highest matching degree with the association attributes of the intervention node are selected based on the association attributes. The matching degree is determined by the correspondence between the association type of the implicit time-series association and the association attributes of the intervention node. Trend associations are prioritized to match the expected impact direction description, conditional associations are prioritized to match the expected association description, and predictive associations are prioritized to match the correction direction description. The selected implicit temporal associations are integrated according to the association type, and the association features that recur in each association type are extracted. The association features include the data node types involved in the association, the temporal relationship of the association, and the node state change features corresponding to the association. The integrated association features are bound to typical occurrence stages to form the core association relationship for each typical occurrence stage. The core association relationship includes the association type, association features and corresponding intervention node association attributes. The core relationships of all typical stages are summarized and arranged in chronological order according to the intervention cycle to form common patterns in the group.

7. A big data-based research data analysis system for cognitive impairment in the elderly, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the big data-based scientific research data analysis method for cognitive impairment in the elderly as described in any one of claims 1 to 6 by executing the machine-executable instructions.

8. A computer program product, characterized in that, The computer program product includes machine-executable instructions stored in a computer-readable storage medium. A processor of a computer device reads the machine-executable instructions from the computer-readable storage medium and executes the machine-executable instructions, causing the computer device to perform the big data-based research data analysis method for cognitive impairment in the elderly as described in any one of claims 1 to 6.

Images