An endocrine patient health management method and system integrated with electronic medical records

Abstract

The present invention relates to the field of intelligent medical technology, and in particular to a health management method and system for endocrine patients integrated with electronic medical records. The method comprises the following steps: collecting multi-dimensional data of endocrine patients through a hospital information system, and constructing a multi-dimensional portrait of the patient to obtain a multi-dimensional health portrait of the patient; extracting time series data features of the multi-dimensional health portrait of the patient to obtain time series feature data; obtaining the patient's historical physiological data; constructing an individualized physiological rhythm model based on the patient's historical physiological data to obtain an individualized physiological rhythm model; and performing non-uniform interpolation on the time series feature data based on the individualized physiological rhythm model to obtain interpolated data. The present invention constructs an individualized physiological rhythm model and a dynamic baseline model, and combines dynamic warning thresholds and subtype stratification to achieve accurate risk assessment and personalized health information management for each patient, thereby significantly improving the effect of health management.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical technology, and in particular to an endocrine patient health management method and system integrated with electronic medical records. Background Art

[0002] Traditional endocrine disease management relies primarily on regular follow-up visits by physicians and patient self-monitoring. This approach has numerous limitations. For example, patients' physiological data is scattered across different medical institutions and systems, making it difficult to integrate and analyze. Physicians spend a significant amount of time collecting and organizing data, which is inefficient. Patients also need to visit the hospital for regular follow-up visits, which is time-consuming and labor-intensive, and can easily lead to decreased compliance. Electronic medical records and information technology were subsequently used to improve the health management of endocrine patients. Early approaches focused on using electronic medical record data for disease risk prediction and early warning. However, these methods were mostly based on population statistics, ignoring individual patient differences and resulting in low accuracy in predictions and early warnings.

[0003] The occurrence and development of endocrine diseases are influenced by multiple factors, including genetics, environment, and lifestyle, resulting in significant differences in disease phenotypes and progression rates among different patients. Health management methods that ignore individual differences have the following drawbacks:

[0004] Inappropriate warning thresholds: Using a unified warning threshold can lead to oversensitivity for some patients, resulting in frequent false alarms, while being insensitive for others, leading to missing real risks. For example, blood sugar fluctuations vary significantly between the elderly and young. Using the same blood sugar warning threshold can result in the elderly receiving frequent hypoglycemia warnings, while the young may not receive timely hyperglycemia warnings.

[0005] Poor health management results: Due to ignoring individual differences among patients, health management plans are difficult to truly meet the needs of patients, resulting in poor management results. Summary of the Invention

[0006] Based on this, it is necessary to provide an endocrine patient health management method and system that integrates electronic medical records to solve at least one of the above technical problems.

[0007] To achieve the above objectives, a method for endocrine patient health management integrated with electronic medical records includes the following steps:

[0008] Step S1: Collect multi-dimensional data of endocrine patients through the integrated electronic medical records in the hospital information system, and construct a multi-dimensional portrait of the patient to obtain a multi-dimensional health portrait of the patient;

[0009] Step S2: extracting time series data features from the patient's multi-dimensional health portrait to obtain time series feature data; obtaining the patient's historical physiological data; constructing an individualized physiological rhythm model based on the patient's historical physiological data to obtain an individualized physiological rhythm model; performing physiological rhythm-based non-uniform interpolation on the time series feature data according to the individualized physiological rhythm model to obtain interpolated data; generating an individualized dynamic baseline model based on the interpolated data to obtain an individualized dynamic baseline model;

[0010] Step S3: Acquire real-time physiological indicator data; calculate physiological indicator deviations based on the individualized dynamic baseline model and the real-time physiological indicator data to obtain physiological indicator deviation values; set dynamic warning thresholds based on the patient's historical physiological data and the individualized dynamic baseline model to obtain warning dynamic thresholds; perform deviation warning identification on physiological indicator deviation values ​​based on the warning dynamic thresholds to obtain a real-time risk assessment report;

[0011] Step S4: Perform standardized feature extraction on the real-time risk assessment report and the patient's multi-dimensional health portrait to obtain standardized feature data; perform DBSCAN cluster analysis on the standardized feature data to obtain clustering results; perform cluster subtype stratification based on the clustering results to obtain subtype feature data; generate an individualized data report based on the subtype feature data and the real-time risk assessment report to obtain an individualized data report to implement endocrine patient health management operations.

[0012] The present invention integrates multi-dimensional data from hospital information systems, wearable devices, and lifestyle questionnaires to construct a comprehensive patient health portrait, laying a solid data foundation for subsequent personalized analysis and precise management, thereby improving the overall effectiveness of health management. By constructing an individualized physiological rhythm model and a dynamic baseline model, personalized predictions and fluctuation range estimates of patient physiological indicators are achieved, providing an important reference basis for subsequent precise early warning and risk assessment, thereby improving the personalization and accuracy of health management. By acquiring physiological indicator data in real time and combining it with an individualized dynamic baseline model and a dynamic early warning threshold, real-time monitoring and deviation early warning of patient physiological indicators are achieved, which can promptly identify potential health risks and provide real-time data support for subsequent risk assessment, thereby improving the timeliness and effectiveness of health management. By performing subtype stratification and cluster analysis on patients and combining it with real-time risk data, personalized data reports are generated, providing patients and medical staff with more intuitive and easier-to-understand health information and personalized management suggestions, thereby improving the pertinence and practicality of health management. Therefore, the present invention provides an endocrine patient health management method integrated with electronic medical records, which effectively solves the shortcomings of existing methods that ignore individual differences among patients. By constructing an individualized physiological rhythm model and a dynamic baseline model, and combining it with dynamic warning thresholds and subtype stratification, this method achieves accurate risk assessment and personalized health information management for each patient, thereby significantly improving the effectiveness of health management.

[0013] Preferably, step S1 includes the following steps:

[0014] Step S11: extracting the electronic medical record data of endocrine patients through the integrated electronic medical record in the hospital information system to obtain the electronic medical record data;

[0015] Step S12: Synchronizing device data of the endocrine patient's wearable medical device through a wireless communication protocol to obtain real-time physiological data;

[0016] Step S13: Acquire lifestyle data; perform data cleaning and preprocessing on the electronic medical record data, real-time physiological data, and lifestyle data to obtain preprocessed patient data;

[0017] Step S14: Construct a multi-dimensional portrait of the patient based on the pre-processed patient data to obtain a multi-dimensional health portrait of the patient.

[0018] By extracting electronic medical record data from hospital information systems, the present invention can comprehensively collect information such as a patient's medical history, diagnosis, treatment, and examinations. This provides the foundational data for constructing a multidimensional patient health profile, preventing information omissions and improving the accuracy of health management. The structured data format also facilitates subsequent data processing and analysis. By synchronizing data with wearable medical devices, patients' physiological indicators, such as blood sugar, heart rate, and blood pressure, can be acquired in real time, promptly reflecting changes in their physiological status, enabling continuous monitoring of patients and improving the timeliness of health management. Acquiring lifestyle data, such as diet, exercise, smoking, and alcohol consumption, can provide a more comprehensive understanding of a patient's health status, identify potential risk factors, and provide more accurate information for personalized health management, thereby improving the effectiveness of health management. By cleaning and preprocessing multi-source data, data quality and consistency can be improved, eliminating the impact of data noise and outliers. Constructing a multidimensional patient health profile integrates static and dynamic patient information to form a comprehensive understanding of their health status, providing a basis for subsequent risk assessment and personalized intervention, ultimately improving the overall effectiveness of health management.

[0019] Preferably, step S2 includes the following steps:

[0020] Step S21: extracting time series data features from the patient's multi-dimensional health portrait to obtain time series feature data; performing time granularity unified processing on the time series feature data to obtain time granularity unified data;

[0021] Step S22: Acquire the patient's historical physiological data; construct an individualized physiological rhythm model based on the patient's historical physiological data to obtain an individualized physiological rhythm model;

[0022] Step S23: performing physiological rhythm-based non-uniform interpolation on the time-granularity unified data according to the individualized physiological rhythm model to obtain interpolated data;

[0023] Step S24: generating time series data based on the interpolated data and the unified time granularity data to obtain time series data; performing time series data segmentation on the time series data to obtain time series segmentation data;

[0024] Step S25: performing recurrent neural network model training on the time series segmentation data to obtain a recurrent neural network model;

[0025] Step S26: inputting the time series data into the recurrent neural network model to predict the fluctuation range of the target physiological indicator to obtain the fluctuation range data of the physiological indicator;

[0026] Step S27: Generate an individualized dynamic baseline model based on the physiological indicator fluctuation range data and the recurrent neural network model to obtain an individualized dynamic baseline model.

[0027] By extracting the characteristics of time series data and unifying the time granularity, the present invention can provide standardized and consistent data input for the subsequent construction of physiological rhythm models and the generation of dynamic baseline models, thereby improving the accuracy and reliability of the models. By constructing an individualized physiological rhythm model, the patient's individual physiological fluctuation patterns, such as postprandial blood sugar fluctuation patterns, circadian rhythms, etc., can be captured, providing a personalized reference basis for the subsequent precise interpolation and dynamic baseline model construction. The non-uniform interpolation method based on the physiological rhythm model can more accurately estimate the missing physiological indicator data, avoid the errors caused by methods such as linear interpolation, and improve the integrity and accuracy of the data, especially in time periods with large fluctuations in physiological indicators, such as after meals. Generating complete time series data and segmenting it can provide standardized input data for the training of the recurrent neural network model, ensuring that the model can learn the temporal dependence and fluctuation patterns of the patient's physiological indicators. By training the recurrent neural network model, the complex temporal patterns of the patient's physiological indicators can be learned, providing a powerful prediction model for the subsequent prediction of the fluctuation range of physiological indicators, thereby improving the accuracy of the prediction. Predicting the fluctuation range of target physiological indicators provides personalized fluctuation range information for the construction of dynamic baseline models, enabling them to better adapt to the individual patient's physiological fluctuation characteristics. By building a personalized dynamic baseline model, baseline levels can be dynamically adjusted according to the individual patient's physiological rhythm and fluctuation range, achieving more accurate deviation warnings and risk assessments, avoiding misjudgments due to individual differences, and improving the personalization and effectiveness of health management.

[0028] Preferably, step S22 includes the following steps:

[0029] Step S221: extracting postprandial blood glucose data from the patient's historical physiological data to obtain a postprandial blood glucose dataset;

[0030] Step S222: performing peak time identification on the postprandial blood glucose dataset to obtain peak time data; performing peak height identification on the postprandial blood glucose dataset to obtain peak height data;

[0031] Step S223: performing peak feature analysis based on the peak time data and the peak height data to obtain peak feature data;

[0032] Step S224: Calculating the blood sugar drop rate for the post-meal blood sugar dataset to obtain drop rate data;

[0033] Step S225: Gaussian process regression is used to construct an individualized physiological rhythm model for the peak characteristic data and the fall-back speed data to obtain an individualized physiological rhythm model.

[0034] By extracting postprandial blood glucose data, the present invention can focus on analyzing the fluctuation of patients' postprandial blood glucose, which is a critical period for blood glucose management and helps to capture the patient's individualized postprandial blood glucose response pattern more accurately. Identifying the postprandial blood glucose peak time and peak height can quantify the key features of the patient's postprandial blood glucose fluctuation and provide basic data for subsequent peak feature analysis. By analyzing the peak time and peak height, we can gain an in-depth understanding of the patient's postprandial blood glucose fluctuation pattern, such as the peak occurrence time and peak amplitude, and provide more refined feature input for the construction of an individualized physiological rhythm model. Calculating the blood glucose drop rate can quantify the speed at which the patient's postprandial blood glucose returns to the baseline level, supplement the postprandial blood glucose fluctuation information other than the peak feature, and enable the physiological rhythm model to more comprehensively describe the patient's postprandial blood glucose response. Using Gaussian process regression to construct an individualized physiological rhythm model can effectively learn and fit the patient's individualized postprandial blood glucose fluctuation pattern, and can make probabilistic predictions of future postprandial blood glucose fluctuations, improve the accuracy and reliability of the prediction, and thus better support personalized health management.

[0035] Preferably, step S23 includes the following steps:

[0036] Step S231: Mark missing values ​​on the unified time granularity data to obtain missing value marked data; locate missing blood glucose values ​​on the missing value marked data to obtain missing blood glucose value location data;

[0037] Step S232: determining the most recent meal time for the missing blood glucose value location data to obtain the most recent meal time data;

[0038] Step S233: Calculating the missing time interval based on the missing blood glucose value location data and the most recent meal time data to obtain time interval data; inputting the most recent meal time data and the time interval data into the individualized circadian rhythm model to predict the blood glucose value to obtain a predicted blood glucose value; using the predicted blood glucose value as a preliminary interpolation of the missing blood glucose value, and generating preliminary interpolated blood glucose data to obtain preliminary interpolated blood glucose data;

[0039] Step S234: extracting exercise and medication information from the patient's multi-dimensional health portrait to obtain exercise and medication information data;

[0040] Step S235: Correcting the preliminary interpolated blood glucose data according to the exercise medication information data to obtain corrected blood glucose data;

[0041] Step S236: performing physiological index interpolation processing of the time-granularity unified data according to the individualized physiological rhythm model, the missing value marked data, and the corrected blood glucose data to obtain interpolated data.

[0042] By marking and locating missing values, the present invention can clearly identify the time points at which blood glucose data require interpolation, providing targeted guidance for the subsequent interpolation process and ensuring its relevance and effectiveness. Determining the time of the most recent meal provides important contextual information for interpolation based on circadian rhythm models, as postprandial blood glucose fluctuations are significantly affected by mealtimes. Using a personalized circadian rhythm model and mealtime information for blood glucose prediction and preliminary interpolation allows for estimation of missing blood glucose values ​​based on individual patient physiology, providing greater accuracy than simple linear interpolation, especially in cases of significant postprandial blood glucose fluctuations. Extracting exercise and medication information identifies key factors influencing blood glucose levels, providing the necessary basis for subsequent correction of interpolation results. Correcting preliminary interpolation results based on exercise and medication information can further improve the accuracy of blood glucose interpolation, as both exercise and medication can significantly affect blood glucose levels. Combining the personalized circadian rhythm model, missing value marking, and corrected blood glucose data for final interpolation generates more complete and accurate time-granular unified data, providing a high-quality data foundation for subsequent dynamic baseline model construction and risk assessment.

[0043] Preferably, step S3 includes the following steps:

[0044] Step S31: acquiring real-time physiological indicator data of the patient through the wearable medical device, and acquiring contextual information related to the indicator, to obtain real-time physiological indicator data and indicator contextual information;

[0045] Step S32: using the individualized dynamic baseline model, performing dynamic baseline model prediction according to the indicator context information to obtain dynamic baseline model prediction data;

[0046] Step S33: performing deviation calculation on the real-time physiological indicator data and the dynamic baseline model prediction data to obtain a physiological indicator deviation value;

[0047] Step S34: setting a dynamic warning threshold based on the patient's historical physiological data and the individualized dynamic baseline model to obtain a warning dynamic threshold; triggering a warning based on the warning dynamic threshold to obtain warning information;

[0048] Step S35: Perform real-time risk assessment based on the early warning information to obtain real-time risk assessment results;

[0049] Step S36: Generate a real-time risk assessment report based on the real-time risk assessment results, warning information, and indicator context information to obtain a real-time risk assessment report.

[0050] By acquiring patient physiological indicator data and contextual information in real time, the present invention can continuously monitor the patient's physiological status and understand the relevant factors that influence changes in physiological indicators, providing real-time, comprehensive data input for subsequent dynamic baseline model predictions and risk assessments. Using a personalized dynamic baseline model and contextual information for predictions, baseline levels can be dynamically adjusted based on the individual patient's physiological characteristics and real-time status, improving the accuracy and personalization of baseline predictions. Calculating the deviation between real-time physiological indicator data and the dynamic baseline predictions quantifies the extent to which physiological indicators deviate from the normal range, providing an important reference indicator for subsequent early warning triggering and risk assessment. Setting dynamic early warning thresholds based on patient historical data and a personalized dynamic baseline model can avoid false positives and false negatives caused by fixed thresholds, improve the accuracy and sensitivity of early warnings, and enable personalized early warning settings based on individual patient conditions. Real-time risk assessments based on early warning information can promptly identify potential health risks. Generating real-time risk assessment reports integrates the patient's real-time physiological data, risk assessment results, and personalized health information, making it easier for medical staff and patients to understand their health status and improving the efficiency and effectiveness of health information management.

[0051] Preferably, step S34 includes the following steps:

[0052] Step S341: extracting the personalized baseline range according to the personalized dynamic baseline model to obtain personalized baseline range data;

[0053] Step S342: Calculating the volatility index of the patient's historical physiological data to obtain volatility index data;

[0054] Step S343: Set the initial threshold value based on the individualized baseline range data to obtain the initial warning threshold value.

[0055] Step S344: dynamically adjusting the initial warning threshold according to the volatility index data to obtain a dynamic warning threshold;

[0056] Step S345: using the early warning dynamic threshold to compare the deviation value of the physiological indicator, and trigger an early warning to obtain early warning information.

[0057] By extracting individualized baseline range data, the present invention can obtain the normal fluctuation range of the patient's physiological indicators, providing a personalized reference basis for the subsequent setting of early warning thresholds. By calculating the volatility index of the patient's historical physiological data, the degree of fluctuation of the patient's physiological indicators can be quantified, providing important reference information for the dynamic adjustment of the early warning threshold. By setting the initial threshold according to the individualized baseline range data, the trigger range of the early warning can be preliminarily determined, and combined with the subsequent dynamic adjustment, a more accurate early warning can be achieved. By dynamically adjusting the early warning threshold according to the volatility index data, the early warning threshold can be better adapted to the individual physiological fluctuation characteristics of the patient, avoiding false alarms and omissions due to individual differences, and improving the accuracy of the early warning. By using the dynamic early warning threshold for deviation comparison and early warning triggering, it is possible to accurately judge whether the early warning needs to be triggered based on the patient's real-time physiological state and individualized fluctuation conditions, thereby improving the timeliness and effectiveness of the early warning.

[0058] Preferably, step S4 includes the following steps:

[0059] Step S41: performing feature selection and extraction on the real-time risk assessment report and the patient's multi-dimensional health portrait to obtain feature data to be clustered;

[0060] Step S42: performing feature dimensionality reduction on the clustered feature data to obtain reduced-dimensional feature data; performing feature standardization on the reduced-dimensional feature data to obtain standardized feature data;

[0061] Step S43: performing DBSCAN cluster analysis on the standardized feature data to obtain clustering results;

[0062] Step S44: performing subtype stratification according to the clustering results to obtain subtype stratification data; performing sub-feature analysis on the subtype stratification data to obtain subtype feature data;

[0063] Step S45: grouping the patients into subtypes according to the subtype stratification data and the subtype characteristic data to obtain patient subtype group data; extracting subtype group features from a preset subtype group characteristic database according to the patient subtype group data to obtain subtype group characteristic data;

[0064] Step S46: extracting real-time risk data from the real-time risk assessment report to obtain real-time risk data;

[0065] Step S47: Generate an individualized data report based on the subtype group characteristic data and the real-time risk data to obtain an individualized data report; perform data visualization processing on the individualized data report to implement health management operations for endocrine patients.

[0066] The present invention can provide effective data input for subsequent cluster analysis by selecting and extracting key features from real-time risk assessment reports and multi-dimensional health portraits of patients, and remove redundant information to improve clustering efficiency and accuracy. Dimensionality reduction and standardization of feature data can reduce data dimensions, eliminate the impact of dimensions between different features, and improve the efficiency and stability of clustering algorithms. Using the DBSCAN algorithm for cluster analysis can effectively identify data clusters of different densities and separate noise points, thereby more accurately stratifying patients into subtypes. Subtype stratification based on clustering results and feature analysis of each subtype can provide an in-depth understanding of the characteristics of patients of different subtypes and provide more detailed information for personalized health management. Patients are grouped according to subtype stratification data and subtype feature data, and corresponding information is extracted from a preset subtype group feature database. This can provide a more targeted management plan for each subtype group and improve the personalization of health management. Extracting real-time risk data can understand the patient's current risk status and provide important real-time information for generating personalized data reports. Combining subtype group characteristic data and real-time risk data to generate personalized data reports and perform data visualization can provide patients and medical staff with more intuitive and easier-to-understand health information, facilitate patient self-management, and assist medical staff in understanding more detailed patient health information, ultimately improving the overall effectiveness of health management.

[0067] Preferably, step S43 includes the following steps:

[0068] Step S431: performing k-nearest neighbor distance calculation on each data point in the standardized feature data to obtain a k-nearest neighbor distance list;

[0069] Step S432: Draw a k-distance graph based on the k-nearest-neighbor distance list to obtain a k-distance graph;

[0070] Step S433: Adaptively determine the neighborhood radius based on the k-distance map to obtain a neighborhood radius list;

[0071] Step S434: Calculating a time decay factor for each data point in the normalized feature data to obtain a time decay factor;

[0072] Step S435: performing time-weighted distance calculation on the standardized feature data according to the time attenuation factor to obtain time-weighted distance data;

[0073] Step S436: Perform DBSCAN cluster analysis on the standardized feature data according to the time-weighted distance data and the neighborhood radius list to obtain clustering results.

[0074] The present invention can quantify the local density of data points in the feature space by calculating the k-nearest neighbor distance of each data point, providing basic data for subsequent drawing of the k-distance graph and determining the neighborhood radius. Drawing the k-distance graph can intuitively display the distribution of the k-nearest neighbor distances of the data points, helping users to more clearly observe changes in data density, thereby more accurately determining the neighborhood radius parameters of the DBSCAN algorithm. Determining the adaptive neighborhood radius based on the k-distance graph can automatically adjust the neighborhood radius parameters according to the actual distribution of the data, avoiding the subjectivity and uncertainty caused by manually setting the parameters, and improving the clustering effect of the DBSCAN algorithm. Calculating the time decay factor can reduce the weight of older data points in the cluster analysis, so that the clustering results are more focused on reflecting the patient's recent situation and improving the timeliness of the clustering results. Calculating the time-weighted distance based on the time decay factor can incorporate the time factor into the distance metric, making the distance calculation more consistent with the actual situation and improving the clustering accuracy of the DBSCAN algorithm. Combining the time-weighted distance and the adaptive neighborhood radius for DBSCAN cluster analysis can comprehensively consider data density and time factors, more accurately identify patient groups of different subtypes, and improve the accuracy and effectiveness of subtype stratification.

[0075] Preferably, the present invention further provides an endocrine patient health management system integrated with electronic medical records, which is used to implement the endocrine patient health management method integrated with electronic medical records as described above. The endocrine patient health management system integrated with electronic medical records includes:

[0076] The multi-dimensional data collection module is used to collect multi-dimensional data of endocrine patients through the integrated electronic medical records in the hospital information system, and to construct a multi-dimensional portrait of the patient to obtain a multi-dimensional health portrait of the patient;

[0077] The time series data modeling module is used to extract time series data features from the patient's multi-dimensional health portrait to obtain time series feature data; obtain the patient's historical physiological data; construct an individualized physiological rhythm model based on the patient's historical physiological data to obtain an individualized physiological rhythm model; perform physiological rhythm-based non-uniform interpolation on the time series feature data based on the individualized physiological rhythm model to obtain interpolated data; generate an individualized dynamic baseline model based on the interpolated data to obtain an individualized dynamic baseline model;

[0078] Deviation warning identification module, used to obtain real-time physiological indicator data; calculate physiological indicator deviation based on the individualized dynamic baseline model and real-time physiological indicator data to obtain physiological indicator deviation value; set dynamic warning threshold based on the patient's historical physiological data and the individualized dynamic baseline model to obtain warning dynamic threshold; perform deviation warning identification on physiological indicator deviation value based on warning dynamic threshold to obtain real-time risk assessment report;

[0079] The subtype hierarchical clustering module is used to extract standardized features from real-time risk assessment reports and multi-dimensional health portraits of patients to obtain standardized feature data; perform DBSCAN clustering analysis on the standardized feature data to obtain clustering results; perform clustering subtype stratification based on the clustering results to obtain subtype feature data; generate individualized data reports based on the subtype feature data and real-time risk assessment reports to obtain individualized data reports, so as to realize endocrine patient health management operations.

[0080] The present invention integrates multi-dimensional data from hospital information systems, wearable devices, and lifestyle questionnaires to construct a comprehensive patient health portrait, laying a solid data foundation for subsequent personalized analysis and precise management, thereby improving the overall effectiveness of health management. By constructing an individualized physiological rhythm model and a dynamic baseline model, personalized predictions and fluctuation range estimates of patient physiological indicators are achieved, providing an important reference basis for subsequent precise early warning and risk assessment, thereby improving the personalization and accuracy of health management. By acquiring physiological indicator data in real time and combining it with an individualized dynamic baseline model and a dynamic early warning threshold, real-time monitoring and deviation early warning of patient physiological indicators are achieved, which can promptly identify potential health risks and provide real-time data support for subsequent risk assessment, thereby improving the timeliness and effectiveness of health management. By performing subtype stratification and cluster analysis on patients and combining it with real-time risk data, personalized data reports are generated, providing patients and medical staff with more intuitive and easier-to-understand health information and personalized management suggestions, thereby improving the pertinence and practicality of health management. Therefore, the present invention provides an endocrine patient health management method integrated with electronic medical records, which effectively solves the shortcomings of existing methods that ignore individual differences among patients. By constructing an individualized physiological rhythm model and a dynamic baseline model, and combining it with dynamic warning thresholds and subtype stratification, this method achieves accurate risk assessment and personalized health information management for each patient, thereby significantly improving the effectiveness of health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 A flowchart of the steps of an endocrine patient health management method integrated with electronic medical records is provided;

[0082] Figure 2 for Figure 1 Detailed implementation steps of step S2 in FIG.

[0083] Figure 3 for Figure 1 Detailed implementation steps of step S3 in FIG.

[0084] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0085] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0086] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0087] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0088] To achieve this, please refer to Figures 1 to 3 , a method for endocrine patient health management integrated with electronic medical records, comprising the following steps:

[0089] Step S1: Collect multi-dimensional data of endocrine patients through the integrated electronic medical records in the hospital information system, and construct a multi-dimensional portrait of the patient to obtain a multi-dimensional health portrait of the patient;

[0090] Step S2: extracting time series data features from the patient's multi-dimensional health portrait to obtain time series feature data; obtaining the patient's historical physiological data; constructing an individualized physiological rhythm model based on the patient's historical physiological data to obtain an individualized physiological rhythm model; performing physiological rhythm-based non-uniform interpolation on the time series feature data according to the individualized physiological rhythm model to obtain interpolated data; generating an individualized dynamic baseline model based on the interpolated data to obtain an individualized dynamic baseline model;

[0091] Step S3: Acquire real-time physiological indicator data; calculate physiological indicator deviations based on the individualized dynamic baseline model and the real-time physiological indicator data to obtain physiological indicator deviation values; set dynamic warning thresholds based on the patient's historical physiological data and the individualized dynamic baseline model to obtain warning dynamic thresholds; perform deviation warning identification on physiological indicator deviation values ​​based on the warning dynamic thresholds to obtain a real-time risk assessment report;

[0092] Step S4: Perform standardized feature extraction on the real-time risk assessment report and the patient's multi-dimensional health portrait to obtain standardized feature data; perform DBSCAN cluster analysis on the standardized feature data to obtain clustering results; perform cluster subtype stratification based on the clustering results to obtain subtype feature data; generate an individualized data report based on the subtype feature data and the real-time risk assessment report to obtain an individualized data report to implement endocrine patient health management operations.

[0093] In the embodiment of the present invention, reference Figure 1 FIG. 1 is a flow chart showing the steps of the endocrine patient health management method integrated with electronic medical records according to the present invention. In this example, the endocrine patient health management method integrated with electronic medical records includes the following steps:

[0094] Step S1: Collect multi-dimensional data of endocrine patients through the integrated electronic medical records in the hospital information system, and construct a multi-dimensional portrait of the patient to obtain a multi-dimensional health portrait of the patient;

[0095] In an embodiment of the present invention, a patient's electronic medical record data, including demographic information, diagnostic information, medication information, test results, and hospitalization information, is extracted through a hospital information system. Simultaneously, the patient's real-time physiological data, such as blood sugar, heart rate, and blood pressure, is synchronized through wearable medical devices. Furthermore, the patient's lifestyle data, such as diet, exercise, smoking, and alcohol consumption, is obtained through questionnaires or mobile phone applications. The collected multidimensional data is cleaned and preprocessed, for example, by removing duplicate values, processing missing values, and performing data standardization. Finally, the preprocessed data is integrated to construct a multidimensional health profile of the patient, for example, using JSON format, which includes dimensions such as demographic characteristics, medical history, physiological indicators, and lifestyle.

[0096] Step S2: extracting time series data features from the patient's multi-dimensional health portrait to obtain time series feature data; obtaining the patient's historical physiological data; constructing an individualized physiological rhythm model based on the patient's historical physiological data to obtain an individualized physiological rhythm model; performing physiological rhythm-based non-uniform interpolation on the time series feature data according to the individualized physiological rhythm model to obtain interpolated data; generating an individualized dynamic baseline model based on the interpolated data to obtain an individualized dynamic baseline model;

[0097] In this embodiment of the present invention, time series data features, such as blood glucose levels, blood pressure, and heart rate, are extracted from a patient's multidimensional health profile and unified to a time granularity, such as hourly. Historical physiological data, such as at least three months of blood glucose data, is obtained. Postprandial blood glucose data is extracted, peak times and peak heights are identified, peak feature analysis is performed, and the rate of blood glucose decline is calculated. Using Gaussian process regression, a personalized circadian rhythm model is constructed using the peak features and decline rate as input. Missing values ​​in the unified time-granularity data are marked and located, the time of the most recent meal is determined, and the missing time interval is calculated. The most recent meal time and time interval are input into the personalized circadian rhythm model to predict and preliminarily interpolate blood glucose values. Exercise and medication information is extracted from the patient's multidimensional health profile, and the preliminary interpolation results are corrected. Finally, the corrected blood glucose values ​​are used to interpolate the unified time-granularity data to obtain complete time series data. This data is used for time series segmentation, and a recurrent neural network model (e.g., LSTM) is trained to predict the fluctuation range of the target physiological indicator. Based on the fluctuation range and the recurrent neural network model, a personalized dynamic baseline model is generated.

[0098] Step S3: Acquire real-time physiological indicator data; calculate physiological indicator deviations based on the individualized dynamic baseline model and the real-time physiological indicator data to obtain physiological indicator deviation values; set dynamic warning thresholds based on the patient's historical physiological data and the individualized dynamic baseline model to obtain warning dynamic thresholds; perform deviation warning identification on physiological indicator deviation values ​​based on the warning dynamic thresholds to obtain a real-time risk assessment report;

[0099] In an embodiment of the present invention, a wearable medical device acquires real-time patient physiological indicator data, such as blood sugar, heart rate, and blood pressure, along with relevant contextual information, such as meal times and exercise types. A personalized dynamic baseline model and indicator contextual information are used to predict the current dynamic baseline value and fluctuation range. The deviation between the real-time physiological indicator data and the predicted dynamic baseline value is calculated. A dynamic warning threshold is set based on the patient's historical physiological data and the personalized dynamic baseline model. Specifically, baseline range data is first extracted from the personalized dynamic baseline model. A volatility indicator (e.g., standard deviation) of the patient's historical physiological data is calculated. An initial warning threshold is set based on the baseline range data, and then the initial threshold is dynamically adjusted based on the volatility indicator. The physiological indicator deviation is compared with the dynamic warning threshold. If the deviation exceeds the threshold, an alert is triggered, generating an alert message. A real-time risk assessment is performed based on the alert message, real-time physiological indicator values, the patient's medical history, and medication use. Finally, the real-time risk assessment results, alert message, and indicator contextual information are integrated to generate a real-time risk assessment report.

[0100] Step S4: extract standardized features from the real-time risk assessment report and the patient's multi-dimensional health profile to obtain standardized feature data; perform DBSCAN cluster analysis on the standardized feature data to obtain clustering results; perform cluster subtype stratification based on the clustering results to obtain subtype feature data; generate an individualized data report based on the subtype feature data and the real-time risk assessment report to obtain an individualized data report to implement endocrine patient health management operations;

[0101] In an embodiment of the present invention, features, such as risk level, demographic characteristics, medical history, physiological indicators, and lifestyle, are extracted from real-time risk assessment reports and multidimensional patient health profiles to form feature data to be clustered. The feature data is dimensionality reduced using the PCA method and normalized using the Z-score method. Cluster analysis is performed on the normalized feature data using the DBSCAN algorithm. The k-nearest neighbor distance of each data point is calculated, a k-distance graph is plotted, and an adaptive neighborhood radius is determined based on the k-distance graph. A time decay factor is calculated for each data point, and a time-weighted distance is calculated based on the time decay factor. DBSCAN clustering is performed using the time-weighted distance and the adaptive neighborhood radius. Subtype stratification is performed based on the clustering results, and characteristic statistical indicators for each subtype are calculated as subtype characteristic data. Patients are divided into different subtype groups, and corresponding subtype group characteristic data is extracted from a pre-set subtype group characteristic database. Real-time risk data is extracted from the real-time risk assessment report. Finally, the subtype group characteristic data and real-time risk data are combined to generate a personalized data report, and data visualization is performed, such as using charts to display the patient's physiological indicators, risk level, and personalized health management information.

[0102] Preferably, step S1 includes the following steps:

[0103] Step S11: extracting the electronic medical record data of endocrine patients through the integrated electronic medical record in the hospital information system to obtain the electronic medical record data;

[0104] Step S12: synchronizing device data of the endocrine patient's wearable medical device through a wireless communication protocol to obtain real-time physiological data;

[0105] Step S13: Acquire lifestyle data; perform data cleaning and preprocessing on the electronic medical record data, real-time physiological data, and lifestyle data to obtain preprocessed patient data;

[0106] Step S14: Construct a multi-dimensional portrait of the patient based on the pre-processed patient data to obtain a multi-dimensional health portrait of the patient.

[0107] In an embodiment of the present invention, the electronic medical record data of endocrine patients is extracted through the database interface of the hospital information system (HIS), for example, using SQL query statements. The extracted data includes but is not limited to: patient basic information (name, age, sex, medical history, family history, etc.), diagnostic information (disease name, diagnosis time, ICD code, etc.), medication information (drug name, dosage, usage, medication time, etc.), test results (blood sugar, blood pressure, blood lipids, etc.), hospitalization information (admission time, discharge time, hospitalization diagnosis, etc.) and other relevant clinical data. The extracted data is stored in a structured data format, such as a CSV file or a database table, for subsequent processing and analysis. For example, for a patient diagnosed with type 2 diabetes, all of his blood sugar records, medication records, and related test results, such as glycated hemoglobin, liver and kidney function, etc., for the past three years are extracted and stored in a CSV file, with each row representing a record and each column representing a data field.

[0108] Data is synchronized with endocrine patients' wearable medical devices (e.g., blood glucose monitors, smartwatches, fitness trackers, etc.) via Bluetooth or other wireless communication protocols. Real-time patient physiological data, such as blood glucose levels, heart rate, blood pressure, sleep duration, and exercise steps, is acquired. A data synchronization frequency is set, for example, every 5 minutes, and the acquired data is timestamped to ensure data consistency. Synchronized data is converted to a new format and stored in a database, for example, using JSON to store data such as blood glucose levels, heart rate, and blood pressure, along with corresponding timestamps. For example, blood glucose data can be automatically synchronized from a patient's continuous glucose monitor (CGM) every 15 minutes and stored in the database as timestamp-blood glucose value pairs, such as "2024-10-27, 10:00:00, 100 mg / dL."

[0109] Obtain the patient's lifestyle data through questionnaires, mobile phone applications or other data collection methods, including but not limited to: eating habits (for example: daily intake of carbohydrates, fats, protein content, meal times, etc.), exercise status (for example: exercise type, exercise duration, exercise intensity, etc.), smoking status, drinking status, sleep quality, etc. Quantify the collected lifestyle data, such as converting eating habits into daily calorie intake and converting exercise status into weekly exercise energy consumption. Store the quantified data as structured data, for example, use numerical variables to record daily carbohydrate intake and use enumeration types to record smoking status (for example: never smoked, quit smoking, still smoking). For example, use a mobile phone APP to record the patient's daily diet, including food types, intake, etc., and use the food database to calculate the daily intake of total calories, carbohydrates, fat and protein ratios.

[0110] Based on the preprocessed patient data obtained in steps S11, S12, and S13, a multidimensional health profile of the patient is constructed. First, the data is cleaned, for example, by removing duplicate values, handling missing values ​​(e.g., using mean-filling or KNN interpolation), and correcting erroneous data. Then, the data is normalized, for example, by converting the data to the same dimension for comparison and analysis. Finally, based on the preprocessed data, a multidimensional health profile of the patient is constructed, including demographic characteristics, medical history, physiological indicators, and lifestyle. For example, information such as the patient's age, gender, BMI, blood glucose history, exercise habits, and dietary habits can be integrated into a unified data structure to form a patient health profile. This profile can be represented, for example, in JSON format, where each dimension contains corresponding specific data. For example, "{'Demographics': {'Age': 50, 'Sex': 'Male'}, 'Medical History': {'Diabetes': 'Type 2'}, 'Physiological Indicators': {'Average Blood Glucose': 150}, 'Lifestyle': {'Exercise': 'Low'}}"

[0111] Preferably, step S2 includes the following steps:

[0112] Step S21: extracting time series data features from the patient's multi-dimensional health portrait to obtain time series feature data; performing time granularity unified processing on the time series feature data to obtain time granularity unified data;

[0113] Step S22: Acquire the patient's historical physiological data; construct an individualized physiological rhythm model based on the patient's historical physiological data to obtain an individualized physiological rhythm model;

[0114] Step S23: performing physiological rhythm-based non-uniform interpolation on the time-granularity unified data according to the individualized physiological rhythm model to obtain interpolated data;

[0115] Step S24: generating time series data based on the interpolated data and the unified time granularity data to obtain time series data; performing time series data segmentation on the time series data to obtain time series segmentation data;

[0116] Step S25: performing recurrent neural network model training on the time series segmentation data to obtain a recurrent neural network model;

[0117] Step S26: inputting the time series data into the recurrent neural network model to predict the fluctuation range of the target physiological indicator to obtain the fluctuation range data of the physiological indicator;

[0118] Step S27: Generate an individualized dynamic baseline model based on the physiological indicator fluctuation range data and the recurrent neural network model to obtain an individualized dynamic baseline model.

[0119] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes:

[0120] Step S21: extracting time series data features from the patient's multi-dimensional health portrait to obtain time series feature data; performing time granularity unified processing on the time series feature data to obtain time granularity unified data;

[0121] In an embodiment of the present invention, time-related physiological indicator data, such as blood glucose values, blood pressure values, heart rates, etc., and time-related events, such as medication time, meal time, exercise time, etc., are extracted from the multi-dimensional health portrait of the patient to form time series feature data. All time series feature data are unified to the same time granularity, for example, once per hour. If the time granularity of the original data is smaller than the target granularity, an aggregation operation is performed, such as calculating the average blood glucose value per hour; if the time granularity of the original data is larger than the target granularity, an interpolation operation is performed, such as using a linear interpolation method to supplement missing values. For example, blood glucose data recorded every 15 minutes is converted into average blood glucose data per hour, and daily medication records are converted into medication dosages per hour (the dosage at the medication time point is the actual dosage, and the dosage at other time points is 0). Finally, data with unified time granularity is obtained.

[0122] Step S22: Acquire the patient's historical physiological data; construct an individualized physiological rhythm model based on the patient's historical physiological data to obtain an individualized physiological rhythm model;

[0123] In an embodiment of the present invention, at least three months of historical physiological data of the patient, such as blood glucose level, heart rate, blood pressure, etc., are obtained. A personalized physiological rhythm model is constructed using the Gaussian process regression method. Taking blood glucose as an example, the historical blood glucose data is used as a training set, time is used as an input variable, and blood glucose level is used as an output variable to train the Gaussian process regression model. The kernel function of the model selects the radial basis function (RBF), and the hyperparameters are optimized using the gradient descent method. Through training, a personalized blood glucose physiological rhythm model of the patient is obtained, which can predict the blood glucose fluctuation trend in the future.

[0124] Step S23: performing physiological rhythm-based non-uniform interpolation on the time-granularity unified data according to the individualized physiological rhythm model to obtain interpolated data;

[0125] In an embodiment of the present invention, according to the personalized physiological rhythm model, non-uniform interpolation based on physiological rhythm is performed on the uniform data of time granularity. For example, if a patient's postprandial blood glucose fluctuates greatly, more intensive interpolation is performed in the postprandial time period, while sparser interpolation is performed in other time periods. The specific interpolation method can use cubic spline interpolation or interpolation based on the predicted value of the physiological rhythm model. For example, for a diabetic patient, according to his personalized blood glucose physiological rhythm model, a blood glucose value is interpolated every 15 minutes within 2 hours after a meal, and a blood glucose value is interpolated every hour in other time periods.

[0126] Step S24: generating time series data based on the interpolated data and the unified time granularity data to obtain time series data; performing time series data segmentation on the time series data to obtain time series segmentation data;

[0127] In this embodiment of the present invention, the interpolated data is merged with the time-granularity unified data to generate complete time series data. The time series data is then segmented according to a fixed time window, for example, using a 24-hour time window as a time window. The time series data is then segmented into multiple data segments of 24 lengths, each representing a day's worth of physiological indicator data, to form time series segmented data. For example, a month's worth of blood glucose data can be segmented into 30 data segments, each containing 24 blood glucose values, based on a 24-hour window.

[0128] Step S25: performing recurrent neural network model training on the time series segmentation data to obtain a recurrent neural network model;

[0129] In an embodiment of the present invention, a recurrent neural network model is trained using time series segmentation data, such as a long short-term memory network (LSTM) or a gated recurrent unit (GRU). Taking blood glucose as an example, the blood glucose data of each time period is used as the input sequence, the blood glucose value at the next time point is used as the output, the mean square error (MSE) is used as the loss function, and the Adam optimizer is used for model training. A recurrent neural network model that can predict future blood glucose values ​​is obtained through training. For example, a recurrent neural network containing two LSTM layers is constructed using the PyTorch framework, with an input dimension of 24 and an output dimension of 1, and the model is trained using three months of historical blood glucose data.

[0130] Step S26: inputting the time series data into the recurrent neural network model to predict the fluctuation range of the target physiological indicator to obtain the fluctuation range data of the physiological indicator;

[0131] In an embodiment of the present invention, time series data is input into a recurrent neural network model to predict the fluctuation range of a target physiological indicator over a period of time. For example, to predict the blood sugar fluctuation range over the next 24 hours, the maximum and minimum hourly blood sugar values ​​can be obtained. For example, by inputting blood sugar data from the past week into a trained LSTM model, the hourly blood sugar fluctuation range for the next 24 hours can be predicted, for example, predicting the blood sugar range at 10 o'clock to be 90-120 mg / dL.

[0132] Step S27: generating an individualized dynamic baseline model based on the physiological indicator fluctuation range data and the recurrent neural network model to obtain an individualized dynamic baseline model;

[0133] In an embodiment of the present invention, an individualized dynamic baseline model is constructed based on the physiological indicator fluctuation range data and a recurrent neural network model. The dynamic baseline model can dynamically adjust the baseline range based on factors such as time and physiological state. For example, the median of the physiological indicator fluctuation range can be used as the dynamic baseline value, and the upper and lower quartiles of the fluctuation range can be used as the fluctuation range of the dynamic baseline. For example, based on the predicted 24-hour blood sugar fluctuation range, the median of the blood sugar per hour is calculated as the dynamic baseline value for that hour, and the upper and lower quartiles of the fluctuation range are calculated as the fluctuation range of the dynamic baseline.

[0134] Preferably, step S22 includes the following steps:

[0135] Step S221: extracting postprandial blood glucose data from the patient's historical physiological data to obtain a postprandial blood glucose dataset;

[0136] Step S222: performing peak time identification on the postprandial blood glucose dataset to obtain peak time data; performing peak height identification on the postprandial blood glucose dataset to obtain peak height data;

[0137] Step S223: performing peak feature analysis based on the peak time data and the peak height data to obtain peak feature data;

[0138] Step S224: Calculating the blood sugar drop rate for the post-meal blood sugar dataset to obtain drop rate data;

[0139] Step S225: Gaussian process regression is used to construct an individualized physiological rhythm model for the peak characteristic data and the fall-back speed data to obtain an individualized physiological rhythm model.

[0140] In an embodiment of the present invention, postprandial blood glucose data is extracted from the patient's historical physiological data to construct a postprandial blood glucose dataset. First, the postprandial time window is determined, for example, 2 hours after a meal. Then, based on the patient's meal records, the blood glucose data within 2 hours after the meal is extracted. If the patient uses a continuous blood glucose monitor (CGM), the blood glucose data of the corresponding time period is directly extracted; if the patient uses a traditional blood glucose meter, the blood glucose data 0.5 hours, 1 hour and 2 hours after the meal are extracted. The extracted postprandial blood glucose data and its corresponding timestamp are stored in the postprandial blood glucose dataset. For example, all blood glucose records marked as "postprandial" for the patient with patient ID 123 in the past three months are queried from the database, and the timestamps and blood glucose values ​​of these records are stored in a new data table to form a postprandial blood glucose dataset.

[0141] For each postprandial blood glucose sequence in the postprandial blood glucose dataset, the peak time and peak height are identified. The peak time is defined as the time point at which the postprandial blood glucose reaches its highest value, and the peak height is defined as the highest value of the postprandial blood glucose. Peaks are identified using a local maximum algorithm. For example, each postprandial blood glucose sequence is traversed to find the local maximum, i.e., the blood glucose value at this time point is greater than the blood glucose values ​​at the two time points before and after it. The time point corresponding to the local maximum is recorded as the peak time, and the local maximum is recorded as the peak height. If there are multiple local maxima, the local maximum with the largest value is selected as the peak.

[0142] Perform peak feature analysis on the peak time data and peak height data. Calculate the difference between the peak time and meal time of each postprandial blood glucose sequence as the peak occurrence time. Calculate the difference between the peak height and the pre-meal blood glucose value as the peak amplitude. Use the peak occurrence time and peak amplitude as peak feature data. For example, calculate the peak occurrence time and peak amplitude of each post-meal blood glucose sequence and store these data in a new data table. For example, a patient's post-meal blood glucose peak occurs 1 hour after the meal, with a peak value of 180 mg / dL, and the pre-meal blood glucose is 90 mg / dL. The peak occurrence time is 1 hour and the peak amplitude is 90 mg / dL.

[0143] The blood glucose drop rate is calculated for each postprandial blood glucose sequence in the postprandial blood glucose dataset. The blood glucose drop rate is defined as the time taken for the peak blood glucose value to drop to the baseline blood glucose value, or as the time taken for the peak blood glucose value to drop to a specific value (such as the pre-meal blood glucose value). A linear regression method can be used to fit a curve of the peak blood glucose value drop, and the slope of the curve can be calculated as the blood glucose drop rate. For example, the least squares method is used to fit a curve of the peak blood glucose value dropping to the pre-meal blood glucose value. The slope of the curve is the blood glucose drop rate, which is expressed in mg / dL / hour.

[0144] A personalized circadian rhythm model is constructed using the Gaussian process regression method. Peak feature data (peak occurrence time and peak amplitude) and fallback speed data are used as input features, and the postprandial blood glucose sequence is used as the output. A radial basis function (RBF) is used as the kernel function, and the hyperparameters are optimized using the gradient descent method. Through training, a personalized circadian rhythm model that can predict postprandial blood glucose fluctuations based on peak features and fallback speed is obtained. For example, the GaussianProcessRegressor function in the Python scikit-learn library is used to train a Gaussian process regression model with peak occurrence time, peak amplitude, and fallback speed as input features and the postprandial blood glucose sequence as the output.

[0145] Preferably, step S23 includes the following steps:

[0146] Step S231: Mark missing values ​​on the unified time granularity data to obtain missing value marked data; locate missing blood glucose values ​​on the missing value marked data to obtain missing blood glucose value location data;

[0147] Step S232: determining the most recent meal time for the missing blood glucose value location data to obtain the most recent meal time data;

[0148] Step S233: Calculating the missing time interval based on the missing blood glucose value location data and the most recent meal time data to obtain time interval data; inputting the most recent meal time data and the time interval data into the individualized circadian rhythm model to predict the blood glucose value to obtain a predicted blood glucose value; using the predicted blood glucose value as a preliminary interpolation of the missing blood glucose value, and generating preliminary interpolated blood glucose data to obtain preliminary interpolated blood glucose data;

[0149] Step S234: extracting exercise and medication information from the patient's multi-dimensional health portrait to obtain exercise and medication information data;

[0150] Step S235: Correcting the preliminary interpolated blood glucose data according to the exercise medication information data to obtain corrected blood glucose data;

[0151] Step S236: performing physiological index interpolation processing of the time-granularity unified data according to the individualized physiological rhythm model, the missing value marked data, and the corrected blood glucose data to obtain interpolated data.

[0152] In this embodiment of the present invention, the unified time-granularity data is traversed, and missing blood glucose values ​​are marked as special values, such as NaN (Not a Number), to generate missing value marker data. The position of the missing blood glucose value in the time series is recorded, such as a timestamp or sequence index, to generate missing blood glucose value location data. For example, in a blood glucose data sequence with an hourly time granularity, if the blood glucose value at 10:00 is missing, the blood glucose value at that time point is marked as NaN, and 10:00 is recorded in the missing blood glucose value location data.

[0153] Based on the missing blood glucose value location, find the most recent mealtime. Obtain mealtime data from the patient's dietary records. Calculate the time difference between each missing value and all mealtimes, and select the mealtime with the smallest time difference as the most recent mealtime. For example, if the missing blood glucose value location is 10:00 AM and the patient's meal records show meals at 7:00 AM and 12:00 PM, the most recent mealtime is 7:00 AM.

[0154] Calculate the time interval between the location of the missing blood glucose value and the time of the most recent meal. Convert the missing value location and the time of the most recent meal into minutes, and then calculate the difference between the two to obtain the time interval data. Input the most recent meal time and time interval data into the personalized circadian rhythm model to predict the blood glucose value at that time point. Use the predicted blood glucose value as a preliminary interpolation of the missing value to generate preliminary interpolated blood glucose data. For example, if the location of the missing blood glucose value is 10:00 and the most recent meal time is 7:00, the time interval is 180 minutes. Input 7:00 and 180 minutes into the personalized circadian rhythm model to obtain a predicted blood glucose value of 120 mg / dL. Use 120 mg / dL as the blood glucose value at 10:00 to generate preliminary interpolated blood glucose data.

[0155] Extract exercise and medication information from the patient's multi-dimensional health profile. Extract exercise type, start time, end time, and intensity from exercise records. Extract medication time, medication name, and dosage from medication records. Combine this extracted information into exercise medication information data. For example, extract that a patient started moderate-intensity running at 9:00 AM, lasted 30 minutes, and took 10 mg of glimepiride at 8:00 AM. Record this information in the exercise medication information data.

[0156] Correct the initially interpolated blood glucose values ​​based on exercise and medication data. For example, if the patient exercised or took glucose-lowering medication during the missing time period, adjust the initially interpolated blood glucose value based on information such as exercise type and intensity, and medication type and dosage. Correction can be performed using pre-defined rules or machine learning-based models. For example, if the patient engaged in moderate-intensity exercise during the missing time period, reduce the initially interpolated blood glucose value by 10 mg / dL.

[0157] Based on the individualized circadian rhythm model, missing value-marked data, and corrected blood glucose data, physiological indicators are interpolated from the uniformly time-granular data. Data points marked as missing are filled with corrected blood glucose values. Data points with non-missing values ​​retain their original values. The resulting interpolated data is then interpolated appropriately based on the circadian rhythm model and exercise medication information. For example, a blood glucose value of 110 mg / dL is added to the blood glucose data at 10:00 AM, while the blood glucose data at other time points remain unchanged, generating the final interpolated data.

[0158] Preferably, step S3 includes the following steps:

[0159] Step S31: acquiring the patient's real-time physiological indicator data and indicator-related context information through the wearable medical device to obtain the real-time physiological indicator data and indicator context information;

[0160] Step S32: using the individualized dynamic baseline model, performing dynamic baseline model prediction according to the indicator context information to obtain dynamic baseline model prediction data;

[0161] Step S33: performing deviation calculation on the real-time physiological indicator data and the dynamic baseline model prediction data to obtain a physiological indicator deviation value;

[0162] Step S34: setting a dynamic warning threshold based on the patient's historical physiological data and the individualized dynamic baseline model to obtain a warning dynamic threshold; triggering a warning based on the warning dynamic threshold to obtain warning information;

[0163] Step S35: Perform real-time risk assessment based on the early warning information to obtain real-time risk assessment results;

[0164] Step S36: Generate a real-time risk assessment report based on the real-time risk assessment results, warning information, and indicator context information to obtain a real-time risk assessment report.

[0165] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes:

[0166] Step S31: acquiring the patient's real-time physiological indicator data and indicator-related context information through the wearable medical device to obtain the real-time physiological indicator data and indicator context information;

[0167] In an embodiment of the present invention, a Bluetooth or other wireless connection is established with the patient's wearable medical device (e.g., a continuous blood glucose monitor, a smart watch) to obtain the patient's physiological indicator data, such as blood glucose level, heart rate, blood pressure, etc. in real time. At the same time, contextual information related to the indicators is obtained, such as: meal time, meal amount, exercise type, exercise duration, medication time, medication dosage, sleep duration, stress level, etc. The acquired physiological indicator data and contextual information are timestamped to ensure the timing and consistency of the data. For example, the blood glucose value is obtained every 5 minutes from the continuous blood glucose monitor worn by the patient, and the meal time, meal amount and other information entered by the patient through the mobile phone APP are recorded at the same time.

[0168] Step S32: using the individualized dynamic baseline model, performing dynamic baseline model prediction according to the indicator context information to obtain dynamic baseline model prediction data;

[0169] In an embodiment of the present invention, a personalized dynamic baseline model is used to predict the dynamic baseline value and fluctuation range of the current time based on the indicator context information. For example, the current time, the time of the most recent meal, the amount of meal, exercise information, etc. are input as input variables into the personalized dynamic baseline model to predict the dynamic baseline value and fluctuation range of blood glucose at the current time. For example, if the current time is 10:00, the patient eats at 8:00 with a meal amount of 50g carbohydrates, and performs 30 minutes of moderate-intensity exercise at 9:00, then this information is input into the dynamic baseline model, and the dynamic baseline value of blood glucose at 10:00 is predicted to be 110mg / dL, with a fluctuation range of 90-130mg / dL.

[0170] Step S33: performing deviation calculation on the real-time physiological indicator data and the dynamic baseline model prediction data to obtain a physiological indicator deviation value;

[0171] In an embodiment of the present invention, a deviation value between the real-time physiological indicator data and the data predicted by the dynamic baseline model is calculated. The deviation value is obtained by subtracting the baseline value predicted by the dynamic baseline model from the real-time physiological indicator data. The deviation value can reflect the degree to which the real-time physiological indicator deviates from the baseline. For example, if the real-time blood glucose value is 140 mg / dL and the baseline value predicted by the dynamic baseline model is 110 mg / dL, the deviation value is 30 mg / dL.

[0172] Step S34: setting a dynamic warning threshold based on the patient's historical physiological data and the individualized dynamic baseline model to obtain a warning dynamic threshold; triggering a warning based on the warning dynamic threshold to obtain warning information;

[0173] In an embodiment of the present invention, a dynamic warning threshold is set based on the patient's historical physiological data and an individualized dynamic baseline model. The dynamic warning threshold can be dynamically adjusted according to the patient's physiological state and historical data. For example, the blood glucose warning threshold can be set to the dynamic baseline value plus or minus a certain proportion of the standard deviation based on the patient's blood glucose fluctuations over the past period of time. For example, if the patient's blood glucose standard deviation over the past week is 20 mg / dL, the blood glucose warning threshold is set to the dynamic baseline value ±2*20 mg / dL. The warning dynamic threshold is compared with the physiological indicator deviation value. If the deviation value exceeds the warning threshold, an early warning is triggered and early warning information is generated, including the warning time, warning type (such as high blood glucose warning or low blood glucose warning), real-time physiological indicator value, dynamic baseline value and warning threshold.

[0174] Step S35: Perform real-time risk assessment based on the early warning information to obtain real-time risk assessment results;

[0175] In embodiments of the present invention, real-time risk assessment is performed based on early warning information. For example, the patient's current health risk level is assessed based on the type of early warning, real-time physiological indicator values, and information such as the patient's medical history and medication use. Risk assessment can be performed using pre-defined rules or machine learning-based models. For example, if a patient triggers a hyperglycemia early warning and their real-time blood sugar level is above 250 mg / dL, the patient's risk level is assessed as high.

[0176] Step S36: Generate a real-time risk assessment report based on the real-time risk assessment results, warning information, and indicator context information to obtain a real-time risk assessment report;

[0177] In an embodiment of the present invention, real-time risk assessment results, early warning information, and indicator context information are integrated to generate a real-time risk assessment report. The report content includes: basic patient information, real-time physiological indicator data, dynamic baseline values, early warning information, risk assessment results, and recommendations and interventions for risks. For example, a real-time risk assessment report is generated that includes the patient's name, age, real-time blood sugar value, dynamic baseline value, high blood sugar early warning information, high-risk assessment results, and a recommendation that the patient immediately take hypoglycemic medication or seek medical attention.

[0178] Preferably, step S34 includes the following steps:

[0179] Step S341: extracting the personalized baseline range according to the personalized dynamic baseline model to obtain personalized baseline range data;

[0180] Step S342: Calculating the volatility index of the patient's historical physiological data to obtain volatility index data;

[0181] Step S343: Set the initial threshold value based on the individualized baseline range data to obtain the initial warning threshold value.

[0182] Step S344: dynamically adjusting the initial warning threshold according to the volatility index data to obtain a dynamic warning threshold;

[0183] Step S345: using the early warning dynamic threshold to compare the deviation value of the physiological indicator, and trigger an early warning to obtain early warning information.

[0184] In an embodiment of the present invention, personalized baseline range data is extracted from a personalized dynamic baseline model. The personalized dynamic baseline model outputs a baseline value and fluctuation range at a specific time point. For example, the baseline blood glucose value at 10:00 AM is 110 mg / dL, with a fluctuation range of 90-130 mg / dL. Baseline values ​​and fluctuation ranges at multiple time points are extracted to form personalized baseline range data. For example, the baseline blood glucose value and fluctuation range for each hour over the next 24 hours are extracted to form a dataset containing 24 baseline values ​​and fluctuation ranges.

[0185] Calculate the volatility of the patient's historical physiological data. Common volatility indicators include standard deviation, coefficient of variation, and glucose excursion. For example, calculate the standard deviation of the patient's average daily glucose levels over the past three months and average these standard deviations to serve as the glucose excursion indicator. Other volatility indicators can also be selected based on specific needs, such as calculating postprandial glucose excursion or nighttime glucose excursion.

[0186] Set the initial warning threshold based on the individualized baseline range data. The initial threshold can be set to the boundary value of the baseline range plus or minus a fixed value or ratio. For example, the initial threshold for blood glucose warning is set to the upper limit of the baseline range plus 15 mg / dL, or to the lower limit of the baseline range minus 15 mg / dL. For example, if the blood glucose baseline range at 10:00 is 90-130 mg / dL, the initial threshold for blood glucose warning at 10:00 is set to 145 mg / dL and 75 mg / dL.

[0187] The initial warning threshold is dynamically adjusted based on the volatility index data. For example, if a patient's blood sugar volatility is high, the warning threshold is set more leniently to avoid frequent warning triggers; if the patient's blood sugar volatility is low, the warning threshold is set more strictly to detect abnormal fluctuations more promptly. For example, the dynamic warning threshold can be set to the initial warning threshold plus or minus a multiple of the volatility index. For example, if a patient's blood sugar volatility index is 20 mg / dL, the dynamic warning threshold is set to the initial warning threshold ± 2 times 20 mg / dL.

[0188] Compare the deviation value of the physiological indicator with the early warning dynamic threshold. If the deviation value exceeds the early warning dynamic threshold, an early warning is triggered and an early warning message is generated. The early warning information includes the early warning time, the early warning type (such as high blood sugar warning or low blood sugar warning), the real-time physiological indicator value, the dynamic baseline value and the early warning threshold. For example, if the real-time blood sugar value at 10:00 is 150mg / dL, the dynamic baseline value is 110mg / dL, and the early warning dynamic thresholds are 145mg / dL and 75mg / dL, then the deviation value is 40mg / dL, which exceeds the early warning dynamic threshold and triggers a high blood sugar early warning. The early warning information includes the early warning time 10:00, the early warning type is high blood sugar warning, the real-time blood sugar value is 150mg / dL, the dynamic baseline value is 110mg / dL, and the early warning threshold is 145mg / dL.

[0189] Preferably, step S4 includes the following steps:

[0190] Step S41: performing feature selection and extraction on the real-time risk assessment report and the patient's multi-dimensional health portrait to obtain feature data to be clustered;

[0191] Step S42: performing feature dimensionality reduction on the clustered feature data to obtain reduced-dimensional feature data; performing feature standardization on the reduced-dimensional feature data to obtain standardized feature data;

[0192] Step S43: performing DBSCAN cluster analysis on the standardized feature data to obtain clustering results;

[0193] Step S44: performing subtype stratification according to the clustering results to obtain subtype stratification data; performing sub-feature analysis on the subtype stratification data to obtain subtype feature data;

[0194] Step S45: grouping the patients into subtypes according to the subtype stratification data and the subtype characteristic data to obtain patient subtype group data; extracting subtype group features from a preset subtype group characteristic database according to the patient subtype group data to obtain subtype group characteristic data;

[0195] Step S46: extracting real-time risk data from the real-time risk assessment report to obtain real-time risk data;

[0196] Step S47: Generate an individualized data report based on the subtype group characteristic data and the real-time risk data to obtain an individualized data report; perform data visualization processing on the individualized data report to implement health management operations for endocrine patients.

[0197] In an embodiment of the present invention, features such as real-time risk level and early warning information are extracted from the real-time risk assessment report. Features such as demographic characteristics (age, gender, BMI, etc.), medical history (type of diabetes, course of disease, etc.), physiological indicators (average blood sugar, blood pressure, blood lipids, etc.), and lifestyle (eating habits, exercise habits, etc.) are extracted from the patient's multi-dimensional health portrait. The extracted features are combined into feature data to be clustered, for example, age, gender, BMI, type of diabetes, average blood sugar, risk level, etc. are combined into a feature vector. All features need to be converted to numerical types, for example, using one-hot encoding to process categorical variables.

[0198] Use principal component analysis (PCA) to reduce the dimensionality of the clustered feature data. A set number of principal components is retained, for example, those that contribute 95% of the variance. The high-dimensional feature data is projected onto a low-dimensional principal component space to obtain reduced-dimensional feature data. The reduced-dimensional feature data is then normalized, for example, using the Z-score method to convert the mean of each feature to 0 and the standard deviation to 1, resulting in standardized feature data.

[0199] Use the DBSCAN algorithm to cluster the standardized feature data. Set the DBSCAN algorithm parameters, such as the neighborhood radius (eps) and the minimum number of samples (min_samples). Use Euclidean distance to calculate the distance between data points. Based on the set parameters, the data points are divided into clusters, along with some noise points. For example, if eps is set to 0.5 and min_samples is set to 5, clustering the standardized feature data using the DBSCAN algorithm will yield three clusters and some noise points.

[0200] Subtype stratification is performed based on the clustering results. Each cluster is considered a subtype. Samples within each subtype are analyzed for characteristics, such as the mean, standard deviation, and median. These statistical indicators are used as subtype feature data to describe the characteristics of each subtype. For example, the mean age, mean blood glucose, and risk level distribution of patients in subtype 1 are calculated as feature data for subtype 1.

[0201] According to the subtype stratification data and subtype characteristic data, the patients are divided into different subtype groups. The characteristic vector of each patient is compared with the characteristic data of each subtype, for example, the distance between the characteristic vector and the subtype center point is calculated, and the patient is divided into the subtype group with the closest distance to obtain the patient subtype grouping data. According to the patient subtype grouping data, the corresponding subtype group characteristic data is extracted from the preset subtype group characteristic database. The subtype group characteristic database contains information such as typical characteristics and risk factors of each subtype. For example, the characteristic vector of patient A is compared with the characteristic data of the three subtypes, and it is found that the distance between patient A and subtype 2 is the closest. Then patient A is divided into subtype 2 group, and the characteristic data of subtype 2 is extracted from the subtype group characteristic database. For example, the typical characteristics of subtype 2 are hyperglycemia, hypertension, and hyperlipidemia, and the risk factors are obesity, lack of exercise, controlling diet, increasing exercise, taking hypoglycemic drugs, etc.

[0202] Extract real-time risk data from the real-time risk assessment report, such as real-time risk level, warning information, and real-time physiological indicator data. For example, extract that the patient's current risk level is high risk, triggering a hyperglycemia warning, and the real-time blood glucose value is 180 mg / dL.

[0203] Combine subtype group characteristic data and real-time risk data to generate personalized data reports. The report content includes: basic patient information, subtype grouping information, real-time risk data, typical characteristics of the subtype group, risk factors, and personalized recommendations and interventions for the patient's current risk. For example, generate a personalized data report that includes the patient's name, age, subtype group (subtype 2), real-time blood sugar value, hyperglycemia warning information, high risk level, typical characteristics and risk factors of subtype 2, and personalized diet, exercise and medication recommendations for the patient's hyperglycemia. Perform data visualization on the personalized data report, such as using charts and graphs to display the patient's physiological indicator data, risk level, and personalized recommendations, so that patients and medical staff can more intuitively understand and use the report content.

[0204] Preferably, step S43 includes the following steps:

[0205] Step S431: performing k-nearest neighbor distance calculation on each data point in the standardized feature data to obtain a k-nearest neighbor distance list;

[0206] Step S432: Draw a k-distance graph based on the k-nearest-neighbor distance list to obtain a k-distance graph;

[0207] Step S433: Adaptively determine the neighborhood radius based on the k-distance map to obtain a neighborhood radius list;

[0208] Step S434: Calculating a time decay factor for each data point in the normalized feature data to obtain a time decay factor;

[0209] Step S435: performing time-weighted distance calculation on the standardized feature data according to the time attenuation factor to obtain time-weighted distance data;

[0210] Step S436: Perform DBSCAN cluster analysis on the standardized feature data according to the time-weighted distance data and the neighborhood radius list to obtain clustering results.

[0211] In an embodiment of the present invention, for each data point in the standardized feature data, the Euclidean distance from the data point to all other data points is calculated. The distances of each data point are sorted in ascending order. The kth distance value after sorting is selected as the k-nearest neighbor distance of the data point. The k-nearest neighbor distances of all data points are stored in a list to obtain a k-nearest neighbor distance list. For example, if the k value is set to 4, the distances from data point A to all other data points are calculated, and the distance list [0.2, 0.3, 0.5, 0.6, 0.7, ...] is obtained after sorting. In this case, the k-nearest neighbor distance of data point A is 0.6.

[0212] Sort the k-nearest-neighbor distance list in ascending order. Plot a k-distance plot using the sorted k-nearest-neighbor distance values ​​as the y-axis and the data point index as the x-axis. The k-distance plot visually shows how the k-nearest-neighbor distances change with each data point.

[0213] Based on the k-distance graph, determine the neighborhood radius (eps) of the DBSCAN algorithm. In the k-distance graph, find the inflection point of the distance curve, that is, the point where the slope of the curve changes significantly. The k-nearest neighbor distance value corresponding to the inflection point is used as the adaptive neighborhood radius. Since the k-nearest neighbor distances of different data points may be different, a different neighborhood radius can be set for each data point to obtain a neighborhood radius list. For example, in the k-distance graph, it is observed that the k-nearest neighbor distance has a clear inflection point near 0.8, so 0.8 is used as the adaptive neighborhood radius.

[0214] Calculate a time decay factor for each data point in the normalized feature data. The time decay factor is used to reduce the weight of older data points in cluster analysis. The time decay factor can be calculated using an exponential decay function: `w(t) = exp(-t / τ)`, where t is the difference between the data point's timestamp and the current time, and τ is the time decay constant. For example, if τ is set to 7 days, for a data point 14 days ago, its time decay factor is exp(-14 / 7) = 0.135.

[0215] A time-weighted distance calculation is performed on the standardized feature data based on the time decay factor. The time-weighted distance is calculated by multiplying the Euclidean distance between data points by the time decay factor. This time-weighted distance can reduce the influence of older data points on the clustering results. For example, if the Euclidean distance between data points A and B is 0.5 and the time decay factor for data point A is 0.135, the time-weighted distance between A and B is 0.5 * 0.135 = 0.0675.

[0216] Use the DBSCAN algorithm to cluster the standardized feature data. Use time-weighted distance as the distance metric and a list of neighborhood radii as the eps parameter. Set the minimum number of samples (min_samples). Based on the time-weighted distance and neighborhood radius, the data points are divided into clusters, along with some noise points. For example, setting min_samples to 5, using time-weighted distance and an adaptive neighborhood radius list, and clustering the standardized feature data using the DBSCAN algorithm yields three clusters and some noise points.

[0217] Preferably, the present invention further provides an endocrine patient health management system integrated with electronic medical records, which is used to implement the endocrine patient health management method integrated with electronic medical records as described above. The endocrine patient health management system integrated with electronic medical records includes:

[0218] The multi-dimensional data collection module is used to collect multi-dimensional data of endocrine patients through the integrated electronic medical records in the hospital information system, and to construct a multi-dimensional portrait of the patient to obtain a multi-dimensional health portrait of the patient;

[0219] The time series data modeling module is used to extract time series data features from the patient's multi-dimensional health portrait to obtain time series feature data; obtain the patient's historical physiological data; construct an individualized physiological rhythm model based on the patient's historical physiological data to obtain an individualized physiological rhythm model; perform physiological rhythm-based non-uniform interpolation on the time series feature data based on the individualized physiological rhythm model to obtain interpolated data; generate an individualized dynamic baseline model based on the interpolated data to obtain an individualized dynamic baseline model;

[0220] Deviation warning identification module, used to obtain real-time physiological indicator data; calculate physiological indicator deviation based on the individualized dynamic baseline model and real-time physiological indicator data to obtain physiological indicator deviation value; set dynamic warning threshold based on the patient's historical physiological data and the individualized dynamic baseline model to obtain warning dynamic threshold; perform deviation warning identification on physiological indicator deviation value based on warning dynamic threshold to obtain real-time risk assessment report;

[0221] The subtype hierarchical clustering module is used to extract standardized features from real-time risk assessment reports and multi-dimensional health portraits of patients to obtain standardized feature data; perform DBSCAN clustering analysis on the standardized feature data to obtain clustering results; perform clustering subtype stratification based on the clustering results to obtain subtype feature data; generate individualized data reports based on the subtype feature data and real-time risk assessment reports to obtain individualized data reports, so as to realize endocrine patient health management operations.

[0222] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0223] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A health management method for endocrine patients integrated with electronic medical records, characterized in that: The following steps are involved: Step S1: Collect multi-dimensional data of endocrine patients through the integrated electronic medical records in the hospital information system, and construct a multi-dimensional portrait of the patient to obtain a multi-dimensional health portrait of the patient; Step S2: extracting time series data features from the patient's multi-dimensional health portrait to obtain time series feature data; Obtain patient historical physiological data; An individualized physiological rhythm model is constructed based on the patient's historical physiological data to obtain an individualized physiological rhythm model; non-uniform interpolation based on physiological rhythm is performed on the time series feature data based on the individualized physiological rhythm model to obtain interpolated data; an individualized dynamic baseline model is generated based on the interpolated data to obtain an individualized dynamic baseline model; Step S3: Acquire real-time physiological indicator data; calculate the physiological indicator deviation based on the individualized dynamic baseline model and the real-time physiological indicator data to obtain the physiological indicator deviation value; set the dynamic warning threshold based on the patient's historical physiological data and the individualized dynamic baseline model to obtain the warning dynamic threshold; Perform deviation warning identification on physiological indicator deviation values ​​according to the warning dynamic threshold and obtain real-time risk assessment report; Step S4: extracting standardized features from the real-time risk assessment report and the patient's multi-dimensional health profile to obtain standardized feature data; Perform DBSCAN cluster analysis on the standardized feature data to obtain clustering results; Cluster subtype stratification is performed based on the clustering results to obtain subtype characteristic data; Generate individualized data reports based on subtype characteristic data and real-time risk assessment reports to obtain individualized data reports to achieve health management operations for endocrine patients.

2. The endocrine patient health management method integrated with electronic medical records according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: extracting the electronic medical record data of endocrine patients through the integrated electronic medical record in the hospital information system to obtain the electronic medical record data; Step S12: Synchronizing device data of the endocrine patient's wearable medical device through a wireless communication protocol to obtain real-time physiological data; Step S13: Acquire lifestyle data; perform data cleaning and preprocessing on the electronic medical record data, real-time physiological data, and lifestyle data to obtain preprocessed patient data; Step S14: Construct a multi-dimensional portrait of the patient based on the pre-processed patient data to obtain a multi-dimensional health portrait of the patient.

3. The endocrine patient health management method integrated with electronic medical records according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: extracting time series data features from the patient's multi-dimensional health portrait to obtain time series feature data; performing time granularity unified processing on the time series feature data to obtain time granularity unified data; Step S22: Acquire the patient's historical physiological data; construct an individualized physiological rhythm model based on the patient's historical physiological data to obtain an individualized physiological rhythm model; Step S23: performing physiological rhythm-based non-uniform interpolation on the time-granularity unified data according to the individualized physiological rhythm model to obtain interpolated data; Step S24: generating time series data based on the interpolated data and the unified time granularity data to obtain time series data; performing time series data segmentation on the time series data to obtain time series segmentation data; Step S25: performing recurrent neural network model training on the time series segmentation data to obtain a recurrent neural network model; Step S26: inputting the time series data into the recurrent neural network model to predict the fluctuation range of the target physiological indicator to obtain the fluctuation range data of the physiological indicator; Step S27: Generate an individualized dynamic baseline model based on the physiological indicator fluctuation range data and the recurrent neural network model to obtain an individualized dynamic baseline model.

4. The endocrine patient health management method integrated with electronic medical records according to claim 3, characterized in that: Step S22 includes the following steps: Step S221: extracting postprandial blood glucose data from the patient's historical physiological data to obtain a postprandial blood glucose dataset; Step S222: performing peak time identification on the postprandial blood glucose dataset to obtain peak time data; performing peak height identification on the postprandial blood glucose dataset to obtain peak height data; Step S223: performing peak feature analysis based on the peak time data and the peak height data to obtain peak feature data; Step S224: Calculating the blood sugar drop rate for the post-meal blood sugar dataset to obtain drop rate data; Step S225: Gaussian process regression is used to construct an individualized physiological rhythm model for the peak characteristic data and the fall-back speed data to obtain an individualized physiological rhythm model.

5. The endocrine patient health management method integrated with electronic medical records according to claim 3 is characterized in that: Step S23 includes the following steps: Step S231: Mark missing values ​​on the unified time granularity data to obtain missing value marked data; locate missing blood glucose values ​​on the missing value marked data to obtain missing blood glucose value location data; Step S232: determining the most recent meal time for the missing blood glucose value location data to obtain the most recent meal time data; Step S233: Calculating the missing time interval based on the missing blood glucose value location data and the most recent meal time data to obtain time interval data; inputting the most recent meal time data and the time interval data into the individualized circadian rhythm model to predict the blood glucose value to obtain a predicted blood glucose value; using the predicted blood glucose value as a preliminary interpolation of the missing blood glucose value, and generating preliminary interpolated blood glucose data to obtain preliminary interpolated blood glucose data; Step S234: extracting exercise and medication information from the patient's multi-dimensional health portrait to obtain exercise and medication information data; Step S235: Correcting the preliminary interpolated blood glucose data according to the exercise medication information data to obtain corrected blood glucose data; Step S236: performing physiological index interpolation processing of the time-granularity unified data according to the individualized physiological rhythm model, the missing value marked data, and the corrected blood glucose data to obtain interpolated data.

6. The endocrine patient health management method integrated with electronic medical records according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: acquiring the patient's real-time physiological indicator data and indicator-related context information through the wearable medical device to obtain the real-time physiological indicator data and indicator context information; Step S32: using the individualized dynamic baseline model, performing dynamic baseline model prediction according to the indicator context information to obtain dynamic baseline model prediction data; Step S33: performing deviation calculation on the real-time physiological indicator data and the dynamic baseline model prediction data to obtain a physiological indicator deviation value; Step S34: setting a dynamic warning threshold based on the patient's historical physiological data and the individualized dynamic baseline model to obtain a warning dynamic threshold; triggering a warning based on the warning dynamic threshold to obtain warning information; Step S35: Perform real-time risk assessment based on the early warning information to obtain real-time risk assessment results; Step S36: Generate a real-time risk assessment report based on the real-time risk assessment results, warning information, and indicator context information to obtain a real-time risk assessment report.

7. The endocrine patient health management method integrated with electronic medical records according to claim 6, characterized in that: Step S34 includes the following steps: Step S341: extracting the personalized baseline range according to the personalized dynamic baseline model to obtain personalized baseline range data; Step S342: Calculating the volatility index of the patient's historical physiological data to obtain volatility index data; Step S343: Set the initial threshold value based on the individualized baseline range data to obtain the initial warning threshold value. Step S344: dynamically adjusting the initial warning threshold according to the volatility index data to obtain a dynamic warning threshold; Step S345: using the early warning dynamic threshold to compare the deviation value of the physiological indicator, and trigger an early warning to obtain early warning information.

8. The endocrine patient health management method integrated with electronic medical records according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: performing feature selection and extraction on the real-time risk assessment report and the patient's multi-dimensional health portrait to obtain feature data to be clustered; Step S42: performing feature dimensionality reduction on the clustered feature data to obtain reduced-dimensional feature data; performing feature standardization on the reduced-dimensional feature data to obtain standardized feature data; Step S43: performing DBSCAN cluster analysis on the standardized feature data to obtain clustering results; Step S44: performing subtype stratification according to the clustering results to obtain subtype stratification data; performing sub-feature analysis on the subtype stratification data to obtain subtype feature data; Step S45: grouping the patients into subtypes according to the subtype stratification data and the subtype characteristic data to obtain patient subtype group data; extracting subtype group features from a preset subtype group characteristic database according to the patient subtype group data to obtain subtype group characteristic data; Step S46: extracting real-time risk data from the real-time risk assessment report to obtain real-time risk data; Step S47: Generate an individualized data report based on the subtype group characteristic data and the real-time risk data to obtain an individualized data report; perform data visualization processing on the individualized data report to implement health management operations for endocrine patients.

9. The endocrine patient health management method integrated with electronic medical records according to claim 8, characterized in that: Step S43 includes the following steps: Step S431: performing k-nearest neighbor distance calculation on each data point in the standardized feature data to obtain a k-nearest neighbor distance list; Step S432: Draw a k-distance graph based on the k-nearest-neighbor distance list to obtain a k-distance graph; Step S433: Adaptively determine the neighborhood radius based on the k-distance map to obtain a neighborhood radius list; Step S434: Calculating a time decay factor for each data point in the normalized feature data to obtain a time decay factor; Step S435: performing time-weighted distance calculation on the standardized feature data according to the time attenuation factor to obtain time-weighted distance data; Step S436: Perform DBSCAN cluster analysis on the standardized feature data according to the time-weighted distance data and the neighborhood radius list to obtain clustering results.

10. An endocrine patient health management system integrated with electronic medical records, characterized in that: For executing the endocrine patient health management method integrated with electronic medical records as claimed in claim 1, the endocrine patient health management system integrated with electronic medical records comprises: The multi-dimensional data collection module is used to collect multi-dimensional data of endocrine patients through the integrated electronic medical records in the hospital information system, and to construct a multi-dimensional portrait of the patient to obtain a multi-dimensional health portrait of the patient; The time series data modeling module is used to extract time series data features from the patient's multi-dimensional health portrait to obtain time series feature data; obtain the patient's historical physiological data; construct an individualized physiological rhythm model based on the patient's historical physiological data to obtain an individualized physiological rhythm model; perform physiological rhythm-based non-uniform interpolation on the time series feature data based on the individualized physiological rhythm model to obtain interpolated data; generate an individualized dynamic baseline model based on the interpolated data to obtain an individualized dynamic baseline model; Deviation warning identification module, used to obtain real-time physiological indicator data; calculate physiological indicator deviation based on the individualized dynamic baseline model and real-time physiological indicator data to obtain physiological indicator deviation value; set dynamic warning threshold based on the patient's historical physiological data and the individualized dynamic baseline model to obtain warning dynamic threshold; perform deviation warning identification on physiological indicator deviation value based on warning dynamic threshold to obtain real-time risk assessment report; The subtype hierarchical clustering module is used to extract standardized features from real-time risk assessment reports and multi-dimensional health portraits of patients to obtain standardized feature data; perform DBSCAN clustering analysis on the standardized feature data to obtain clustering results; perform clustering subtype stratification based on the clustering results to obtain subtype feature data; generate individualized data reports based on the subtype feature data and real-time risk assessment reports to obtain individualized data reports, so as to realize endocrine patient health management operations.

Images