Identification and early warning method and device for irritable hyperglycemia, electronic equipment and storage medium

Abstract

The invention provides a recognition and early warning method and device for irritable hyperglycemia, electronic equipment and a storage medium, and is applied to the technical field of medical data processing. According to the method, data type information (including FreeStyle LibreH dynamic blood glucose and glycosylated hemoglobin data), influence factor data (improved and preset irritable hyperglycemia index and the like) and abnormal state information (fasting interference and detection difference) are taken; acute-stage and chronic-stage average blood glucose basic data are generated through preprocessing. The index weight is enhanced and improved by means of a dynamic fusion engine, the probability is predicted by combining space-time correlation analysis and binary logistic regression, and the abnormal characteristics are obtained by comparing the difference of the AUC by adopting Delog test. The multi-modal features are processed to obtain an identification result, clinical indexes screened by LASSO are fused, index differences are compared to generate risk influence factors, finally, a dynamic early warning result is generated based on a space-time correlation early warning engine and a multi-target strategy, and accurate identification and early warning are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical data processing, and particularly relates to a stress hyperglycemia identification and early warning method and device, electronic equipment and storage medium. BACKGROUND

[0002] The most commonly used technology for judging stress hyperglycemia is to evaluate it through a stress hyperglycemia index, which is a ratio, and the denominator of the ratio is the random blood glucose on admission. The accuracy of the result is affected by the uncertain fasting time of the patient. For example, two patients have the same degree of blood glucose fluctuation, but the fasting time and food intake of the two patients before measuring the random blood glucose on admission are completely different, and the measured blood glucose values are also completely different. This non-random error has a greater impact on patients with cardiovascular acute and critical conditions, such as patients with acute and severe conditions who go to the emergency department for treatment. The blood glucose value measured at the time of going to the emergency department is not a fasting blood glucose value, and whether it is fasting or not significantly affects the blood glucose level, thereby causing the prior art to be significantly affected by the fasting time and food intake before the onset of the disease.

[0003] The most commonly used technology for judging stress hyperglycemia is to evaluate it through a stress hyperglycemia index, which is a ratio, and the denominator of the ratio is the random blood glucose on admission. The accuracy of the result is affected by the uncertain fasting time of the patient. For example, two patients have the same degree of blood glucose fluctuation, but the fasting time and food intake of the two patients before measuring the random blood glucose on admission are completely different, and the measured blood glucose values are also completely different. This non-random error has a greater impact on patients with cardiovascular acute and critical conditions, such as patients with acute and severe conditions who go to the emergency department for treatment. The blood glucose value measured at the time of going to the emergency department is not a fasting blood glucose value, and whether it is fasting or not significantly affects the blood glucose level, thereby causing the prior art to be significantly affected by the fasting time and food intake before the onset of the disease.

[0004] It should be noted that the information disclosed in the above background section is only used to strengthen the understanding of the background of the present disclosure, and therefore can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] According to one aspect of this application, a method for identifying and providing early warning of stress-induced hyperglycemia is provided, comprising: acquiring data type information, influencing factor data, and abnormal state information; performing data preprocessing on the data type information to generate basic data on average blood glucose in the acute phase and average blood glucose in the chronic phase; processing the influencing factor data and abnormal state information based on a dynamic fusion engine, strengthening the core influence weight of the modified stress-induced hyperglycemia index through a dynamic weight allocator, calculating the predictive probability of the modified index for adverse events using a binary logistic regression model in conjunction with spatiotemporal correlation analysis, comparing the AUC difference between the modified index and the preset index using the Delong test, and generating abnormal feature identification information of the modified stress-induced hyperglycemia index; and performing multi-... Modal dynamic feature information is processed to generate stress-induced hyperglycemia identification results. The identification results and risk levels in the influencing factor data are combined with a multi-task decision matrix. The stress-induced hyperglycemia identification results corresponding to the modified stress-induced hyperglycemia index and the cardiovascular adverse event risk levels are combined with the multi-task decision matrix, and clinical indicators independently associated with adverse prognosis selected by LASSO regression are integrated. The risk prediction difference between the preset index and the modified index is compared to generate the risk influencing factors of stress-induced hyperglycemia on cardiovascular adverse events. Based on the spatiotemporal correlation early warning engine combined with a multi-objective dynamic decision strategy, the identification results, risk influencing factors and abnormal feature identification information are processed to generate dynamic early warning results for stress-induced hyperglycemia.

[0006] Another aspect of this application discloses a device for identifying and warning of stress-induced hyperglycemia, comprising: an acquisition module for acquiring data type information, influencing factor data, and abnormal state information; a processing module for preprocessing the data type information to generate basic data on average blood glucose in the acute and chronic phases; processing the influencing factor data and abnormal state information based on a dynamic fusion engine, strengthening the core influence weights of the modified stress-induced hyperglycemia index through a dynamic weight allocator, calculating the predictive probability of adverse events by the modified index using a binary logistic regression model in conjunction with spatiotemporal correlation analysis, comparing the AUC difference between the modified index and the preset index using the Delong test, and generating abnormal feature identification of the modified stress-induced hyperglycemia index. The system analyzes various aspects of hyperglycemia. It processes multimodal dynamic feature information to generate stress-induced hyperglycemia identification results. It combines the identification results and risk levels in the influencing factor data with a multi-task decision matrix. The modified stress-induced hyperglycemia index, corresponding to the stress-induced hyperglycemia identification results and cardiovascular adverse event risk levels, are integrated with the multi-task decision matrix. Clinical indicators independently associated with adverse prognosis, selected through LASSO regression, are fused. The risk prediction differences between the preset index and the modified index are compared to generate risk influencing factors for stress-induced hyperglycemia on cardiovascular adverse events. Finally, based on a spatiotemporal correlation early warning engine combined with a multi-objective dynamic decision strategy, the identification results, risk influencing factors, and abnormal feature identification information are processed to generate dynamic early warning results for stress-induced hyperglycemia.

[0007] According to another aspect of this application, an electronic device includes: a first processor; and a memory for storing executable instructions of the first processor; wherein the first processor is configured to execute the above-described method for identifying and warning of stress-induced hyperglycemia by executing the executable instructions.

[0008] According to another aspect of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a second processor, implements the above-described method for identifying and warning of stress-induced hyperglycemia.

[0009] This application provides a method, device, electronic device, and storage medium for identifying and providing early warning of stress-induced hyperglycemia. It acquires FreeStyle LibreH continuous glucose monitoring data (14-day continuous monitoring) and glycated hemoglobin data detected by a Tosoh G8 HPLC analyzer, and preprocesses the data to generate average blood glucose levels for the acute and chronic phases. A dynamic fusion engine is used to enhance and improve the weighting of the stress-induced hyperglycemia index. Spatiotemporal correlation analysis and binary logistic regression are combined to calculate the predicted probability, and the AUC advantage is verified based on the Delong test to generate abnormal feature information. Multimodal features are processed to obtain identification results, and clinical indicators selected by LASSO are fused. Risk influencing factors are generated by comparing index differences. Finally, based on a spatiotemporal correlation early warning engine and a multi-objective strategy, dynamic early warning results are generated, achieving accurate identification of stress-induced hyperglycemia and early warning of cardiovascular adverse event risks, improving the accuracy of clinical assessment and the timeliness of intervention.

[0010] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0011] Figure 1 A flowchart illustrating a method for identifying and providing early warning of stress-induced hyperglycemia according to an embodiment of this application is shown. Figure 2 This illustration shows a schematic diagram of the structure of a stress-induced hyperglycemia identification and early warning device provided in an embodiment of this application. Detailed Implementation

[0012] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0013] The following is combined with Figure 1This application describes a method for identifying and providing early warning of stress-induced hyperglycemia according to an exemplary embodiment. It should be noted that the following application scenarios are shown only to facilitate understanding of the spirit and principles of this application, and the embodiments of this application are not limited in any way. Rather, the embodiments of this application can be applied to any applicable scenario.

[0014] In one implementation, Figure 1 A schematic flowchart illustrating a method for identifying and providing early warning of stress-induced hyperglycemia according to an embodiment of this application is shown.

[0015] S101, obtain data type information, impact factor data, and abnormal status information.

[0016] In one implementation, the data type information serves as the foundation for subsequent calculations of average blood glucose during the acute phase, average blood glucose during the chronic phase, and the modified stress hyperglycemic index. It is necessary to clearly define the data source, acquisition equipment, detection indicators, and data characteristics to ensure data traceability and reproducibility. The FreeStyle LibreH continuous glucose monitoring device (Abbott, Chicago, American) was selected. This device monitors blood glucose through a subcutaneously worn sensor that can be worn continuously for 14 days, automatically measuring and storing interstitial fluid glucose data. Healthcare personnel can download data from the sensor at any time (whether or not it is worn), eliminating the need for frequent blood draws. This is suitable for clinical scenarios where cardiovascular patients (especially those with acute and critical illnesses) cannot undergo frequent blood draws.

[0017] The collected indicators include continuous glucose values ​​during the patient's acute phase. The device can output multi-dimensional derived indicators. Taking Chen Yang, a cardiology patient (test date August 14, 2025, report number 11000294), as an example, the device collected 4.6 days of valid blood glucose data (100% valid data ratio), and output the following indicators: average blood glucose 6.33 mmol / L, glucose management index (GMI) 6%, coefficient of variation (CV) 24.6%, standard deviation (SD) 1.56 mmol / L, average blood glucose fluctuation range 3.57 mmol / L, average absolute difference in blood glucose during the day 1.29 mmol / L, percentage of blood glucose within the target range (TIR) ​​94.2%, percentage of hyperglycemic time (TAR) 4%, and percentage of hypoglycemic time (TBR) 1.8%. These indicators can comprehensively reflect the patient's blood glucose level and fluctuation during the acute phase, providing data support for calculating the average blood glucose during the acute phase.

[0018] Blood samples were obtained from patients via venous blood collection using a Tosoh G8 HPLC analyzer. This device, based on high-performance liquid chromatography (HPLC), detects glycated hemoglobin (HbA1c) levels with high accuracy, accurately reflecting the patient's average blood glucose control level over the past 2-3 months and avoiding the random errors of single blood glucose tests. The collected indicator was the patient's HbA1c percentage. For Mr. Li, a patient diagnosed with coronary heart disease, venous blood was collected, and the HbA1c value was measured at 6.5% using the Tosoh G8 HPLC analyzer. This data will be used to estimate the average blood glucose level during the chronic phase and is one of the key parameters for calculating the modified stress-induced hyperglycemia index. Simultaneously, the error range during the testing process needs to be recorded. The error in HbA1c detection using this device is typically controlled within ±0.1%, ensuring data accuracy.

[0019] Impact factor data are key variables for assessing the association between stress-induced hyperglycemia and adverse cardiovascular events. They must cover core indices, event outcomes, and potential association indicators, clearly defining the definitions, calculation methods, and data sources of each factor. Modified Stress-Induced Hyperglycemia Index (baseline data to be calculated): The core calculation parameters for this index are "average blood glucose in the acute phase" and "average blood glucose in the chronic phase." These need to be further calculated based on the dynamic blood glucose monitoring data and glycated hemoglobin data from the aforementioned data types. Currently, the raw baseline data required for calculating this index is available, as shown in the example below: Patient Wang's average blood glucose in the acute phase, monitored by a FreeStyleLibreH device, was 7.2 mmol / L; glycated hemoglobin, measured by a Tosoh G8 HPLC analyzer, was 7.0%. Substituting these values ​​into the formula "average blood glucose in the chronic phase (mg / dL) = (28.7 × HbA1c%) - 46.7", the average blood glucose in the chronic phase can be estimated as (28.7 × 7.0) - 46.7 = 154.2 mg / dL (the modified index needs to be calculated after unifying the units).

[0020] The preset stress-induced hyperglycemia index is calculated using "random blood glucose upon admission" and "estimated average blood glucose during the chronic phase" as parameters. It requires obtaining the random blood glucose data upon admission and the corresponding glycated hemoglobin (HbA1c) data. For example: Patient Zhang had venous blood collected upon admission to measure random blood glucose, which was 180 mg / dL. Simultaneously, HbA1c was measured at 6.8%. Substituting these values ​​into the formula, the estimated average blood glucose during the chronic phase is (28.7 × 6.8) - 46.7 = 148.46 mg / dL. Using the existing technical formula, the preset index can be calculated as 180 ÷ 148.46 ≈ 1.21. This data will be used to compare the AUC difference with the modified index later.

[0021] Adverse cardiovascular events include acute myocardial infarction, exacerbation of heart failure, malignant arrhythmias, and cardiac death. It is necessary to collect outcome information on whether these events occurred during the patient's future follow-up period (usually 6 months to 1 year). Data sources include hospital electronic medical records, outpatient follow-up records, and telephone follow-up records. For example, a cardiovascular patient, Mr. Zhao, admitted in January 2024, was followed up for one year. His emergency room medical record from October 2024 showed that he was readmitted due to "acute myocardial infarction," which was considered an adverse cardiovascular event. Patient Mr. Sun did not have any readmissions for cardiovascular disease within one year of follow-up, and his outpatient electrocardiogram and echocardiogram showed no significant abnormalities, thus he was considered to have not experienced any adverse events. These outcome data will be used as the dependent variable in the subsequent binary logistic regression model.

[0022] Potentially relevant clinical indicators include patients' basic clinical information (age, gender, body mass index), history of underlying diseases (history of hypertension, history of diabetes, history of hyperlipidemia), laboratory test indicators (four lipid parameters, liver and kidney function, myocardial enzyme profile), and treatment regimen (whether insulin, beta-blockers, etc. are used). These indicators may be related to poor prognosis and need to be collected as potential influencing factors to provide data for subsequent LASSO regression screening of independently relevant indicators.

[0023] Abnormal status information is crucial for identifying defects in existing technologies and optimizing technical solutions. It is necessary to clearly define the specific scenarios and manifestations that lead to data errors in existing technologies to ensure that subsequent technical solutions can specifically avoid these anomalies. The denominator of the preset stress-induced hyperglycemia index is the random blood glucose level upon admission. However, random blood glucose levels can vary significantly depending on the length of fasting and the amount of food consumed before admission, leading to deviations in the index calculation results. Therefore, it is necessary to collect random blood glucose data upon admission under different fasting times and food intakes, along with the corresponding error manifestations. Patients Chen and Lin were both admitted with acute coronary syndrome, and their glycated hemoglobin levels upon admission were both 6.5% (estimated average chronic phase blood glucose levels were the same): Patient Chen fasted for 8 hours before admission (fasting state), and his random blood glucose level upon admission was 110 mg / dL; Patient Lin ate 200g of rice 1 hour before admission (not fasting state), and his random blood glucose level upon admission was 160 mg / dL. The actual blood glucose fluctuations of the two individuals were similar, but due to the difference in fasting time, the random blood glucose difference upon admission reached 50 mg / dL. If the preset index is calculated using the existing technical formula, Chen's index is 110 ÷ [(28.7 × 6.5) - 46.7] ≈ 110 ÷ 140.85 ≈ 0.78, while Lin's index is 160 ÷ 140.85 ≈ 1.14. The significant difference between the two indices directly reflects the interference of fasting time on the results of the existing technology.

[0024] The numerator of the pre-defined stress-induced hyperglycemia index, "estimated average blood glucose in the chronic phase," is calculated based on the interstitial fluid glucose level measured by a continuous glucose monitoring device. The denominator, "random blood glucose upon admission," is the venous blood glucose level. The two have fundamental differences in their detection methods (interstitial fluid glucose vs. venous blood glucose) and detection sites (subcutaneous tissue vs. veins), leading to an imprecise ratio calculation. It is necessary to collect interstitial fluid glucose data and venous blood glucose data from the same patient during the same period to clarify the differences between the two.

[0025] S102, perform data preprocessing on data type information to generate basic data on average blood glucose in the acute phase and average blood glucose in the chronic phase.

[0026] In one implementation, based on clinical needs for accurate blood glucose data and considering interference from fasting time and differences in detection methods, the blood glucose monitoring device in the data type information is selected and its parameters are determined. This yields 14 days of continuous monitoring parameters and accurate glycated hemoglobin (HbA1c) detection parameters from the FreeStyle LibreH dynamic blood glucose monitoring device. The device sensor is worn for 14 days, automatically collecting interstitial fluid glucose data every 15 minutes, generating 96 data points per day to ensure data continuity. The data storage capacity supports 14 days of complete data storage, and medical staff can download sensor data at any time via dedicated equipment (regardless of whether it is worn), ensuring data traceability. The detection range covers 3.9-21.0 mmol / L, meeting the monitoring needs of cardiovascular patients during acute blood glucose fluctuations and avoiding data loss due to insufficient detection range. For Wang, a patient admitted to the hospital with acute myocardial infarction, a FreeStyle LibreH sensor was fitted on the day of admission, and the monitoring period was set to 14 days. The device automatically collected blood glucose data at a frequency of 15 minutes / time. On the 7th day, medical staff downloaded the data of the previous 7 days through a dedicated reader, obtaining a total of 7×96=672 valid blood glucose data points without any data interruption, providing a continuous data source for subsequent calculation of the average blood glucose during the acute phase.

[0027] The test sample type is venous whole blood, and the sample volume is only 5μL, which is suitable for routine venous blood collection in cardiovascular patients; the test time is 1.5 minutes per sample, which can quickly obtain results and meet the needs of emergency clinical diagnosis and treatment; the test error range is controlled within ±0.1%, ensuring the accuracy of glycated hemoglobin data and providing reliable parameters for subsequent estimation of average blood glucose in the chronic phase.

[0028] Based on the priority of data usage, the data collected by the two types of devices were initially screened. Priority was given to ensuring the continuity of dynamic blood glucose data required for calculating the average blood glucose in the acute phase and the accuracy of glycated hemoglobin data, while also ensuring the completeness of average blood glucose, coefficient of variation, and blood glucose fluctuation range in the dynamic blood glucose data. The core screening objective was to ensure the continuity of data required for calculating the average blood glucose in the acute phase, and the secondary objective was to ensure the completeness of auxiliary indicators such as average blood glucose, coefficient of variation, and blood glucose fluctuation range. The screening rules were set as follows: First, the data continuity requirement was that the percentage of valid data points per day was ≥90% (i.e., ≤9.6 missing data points per day, rounded up to ≤9). If more than 9 missing data points were found in a single day, the data for that day would be discarded, and only valid data days would be retained. Second, the data rationality requirement was to remove outliers that exceeded the device's detection range (3.9-21.0 mmol / L). If the percentage of outliers in a single day was ≥10%, the retention of the data for that day would be determined based on clinical judgment (e.g., whether the patient experienced hypoglycemia / hyperglycemia symptoms).

[0029] The core screening objective is to ensure the accuracy of the data required for estimating the average blood glucose level in the chronic phase. The screening rules are as follows: First, the test report must be complete, including the test result, test time, sample number, tester's signature, and equipment calibration record. No information can be missing. Second, test error verification: if the same sample is tested twice in parallel, the difference between the two results must be ≤0.1%. If the difference exceeds 0.1%, the sample must be collected and tested again. Third, clinical rationality verification: the glycated hemoglobin test result must be within the clinically reasonable range (generally 4.0%-10.0%). If it exceeds this range, it is necessary to determine whether the result is true based on the patient's past medical history (such as whether the patient has diabetes).

[0030] Constraints on data logic consistency, data integrity, and adaptation to the cardiovascular patient monitoring scenario are applied. The average blood glucose is calculated from the screened dynamic blood glucose data, and the chronic phase average blood glucose is estimated from the glycated hemoglobin data. The formula for chronic phase average blood glucose is: (28.7 × HbA1c%) - 46.7. For data logic consistency, the unit of dynamic blood glucose data must be unified to mmol / L (the original unit of the device data is mmol / L, no conversion is needed) to avoid unit confusion. For data integrity, the effective data days used to calculate the average blood glucose must be ≥7 days (if the patient's acute phase is less than 14 days, it is calculated based on the actual number of days of hospitalization, and the effective data days must be ≥80% of the number of days of hospitalization). For scenario adaptation, the acute phase for cardiovascular patients is usually defined as 7-14 days after admission; here, the average blood glucose for the acute phase is calculated based on 14 days after admission (the effective data days after screening). If the patient's hospitalization is less than 14 days, the calculation is based on the effective data from the actual number of days of hospitalization.

[0031] The average blood glucose level during the acute phase is calculated as follows: (sum of the daily average blood glucose levels for all valid data days after screening) ÷ number of valid data days; where the daily average blood glucose level is calculated as: (sum of valid blood glucose data points for that day) ÷ number of valid data points for that day. Patient Wang retained 13 days of valid data after screening. The daily average blood glucose levels for each valid data day were 7.2, 7.0, 6.9, 7.1, 7.3, 7.2, 7.0, 6.8, 6.9, 7.1, 7.0, 6.9, and 7.2 mmol / L. The sum of these 13 daily average blood glucose levels is 91.6 mmol / L. Therefore, the average blood glucose level during the acute phase is approximately 91.6 ÷ 13 ≈ 7.05 mmol / L. The calculation result is rounded to two decimal places, which meets the accuracy requirements for clinical data.

[0032] Data logic consistency constraint: Glycated hemoglobin data (percentage) must be substituted into a unified formula for calculation; data integrity constraint: Valid glycated hemoglobin data after screening must be used, and abnormal data that has not passed screening cannot be used; scenario adaptability constraint: The unit of the formula calculation result is mg / dL, which must be consistent with the unit of the average blood glucose in the acute phase (mmol / L) in the future to avoid logical contradictions caused by unit differences.

[0033] The chronic phase average blood glucose was calculated using the formula: Chronic phase average blood glucose (mg / dL) = (28.7 × HbA1c%) - 46.7. To ensure consistency with the acute phase average blood glucose (mmol / L), unit conversion is required: 1 mmol / L = 18 mg / dL, i.e., chronic phase average blood glucose (mmol / L) = chronic phase average blood glucose (mg / dL) ÷ 18. Patient Wang's screened glycated hemoglobin level was 6.8%. Substituting this into the formula, the chronic phase average blood glucose (mg / dL) was calculated as: Chronic phase average blood glucose (mg / dL) = (28.7 × 6.8) - 46.7 = 195.16 - 46.7 = 148.46 mg / dL. Converting to mmol / L: 148.46 ÷ 18 ≈ 8.25 mmol / L. This result will serve as the basis for subsequent comparisons with the acute phase average blood glucose.

[0034] The integrated acute-phase and chronic-phase average blood glucose data are used to generate preprocessed baseline data for both. This data includes quality verification information such as the percentage of valid data from continuous glucose monitoring (CGM) and the error range for glycated hemoglobin (HbA1c) testing. The integrated data includes basic patient information (name, gender, age, admission date, clinical diagnosis), acute-phase average blood glucose data (calculation results, number of valid data days, data source device number), and chronic-phase average blood glucose data (estimated results, raw HbA1c data, data source device number), ensuring a one-to-one correspondence between data and patients to avoid confusion. The primary quality verification indicator for CGM is the percentage of valid data days, calculated as: Valid data percentage = (Number of valid data days ÷ Total monitoring days) × 100%. If the valid data percentage is ≥80%, the data quality is considered acceptable; if it is below 80%, the monitoring protocol needs to be re-evaluated.

[0035] The main verification indicator for glycated hemoglobin (HbA1c) testing quality is the testing error range, i.e., the error value indicated on the test report. If the error range is ≤ ±0.1%, the data quality is considered acceptable; if it exceeds this range, retesting is required. Patient Wang's total monitoring days were 14 days, with 13 days of valid data. The percentage of valid data = (13 ÷ 14) × 100% ≈ 92.86% ≥ 80%, indicating that the dynamic blood glucose monitoring data quality is acceptable. The HbA1c testing error range is ±0.08% ≤ ±0.1%, indicating that the testing data quality is acceptable. The above quality verification information is attached to the baseline blood glucose data to form a complete baseline data document, marked "Data quality acceptable, can be used for subsequent analysis."

[0036] S103 processes the impact factor data and abnormal state information based on the dynamic fusion engine. It strengthens the core impact weight of the modified stress hyperglycemia index through a dynamic weight allocator. Combined with spatiotemporal correlation analysis, it uses a binary logistic regression model to calculate the prediction probability of adverse events by the modified index. Combined with the Delong test, it compares the AUC difference between the modified index and the preset index to generate abnormal feature identification information of the modified stress hyperglycemia index.

[0037] In one implementation, weight allocation rules are defined and interference factors are prioritized for the modified stress-induced hyperglycemia index, the preset stress-induced hyperglycemia index, and the fasting time interference and detection method difference data in the abnormal state information from the influencing factor data. This generates the core weight parameters of the modified index, the abnormal state interference coefficient, and the weight decay ratio of the preset index, forming the basic information for index weight allocation. The modified stress-induced hyperglycemia index is used as the core influencing factor. Based on its technical advantages in avoiding "fasting time interference" and "detection method differences," its core weight parameter is set to 0.7 (70% weight ratio) to ensure its dominant role in the subsequent prediction model. Due to two major technical defects, the preset stress-induced hyperglycemia index has a weight decay ratio of 0.3 (i.e., only 30% of the original weight is retained; the actual weight ratio = original base weight × 0.3; if the original base weight is set to 0.5, the decayed weight ratio is 0.15), weakening its interference with the results.

[0038] Interference coefficients were set for "fasting time interference" and "difference in testing methods" in the abnormal status information. Fasting time interference had a more significant impact on the preset index (e.g., differences in fasting time can lead to a deviation of up to 50 mg / dL in blood glucose levels), so an interference coefficient of 0.4 was set for it. Differences in testing methods (difference between interstitial fluid and venous blood glucose) had a relatively smaller impact (a deviation of approximately 15 mg / dL), so an interference coefficient of 0.2 was set for it. A larger interference coefficient indicates a stronger negative impact of that factor on data accuracy, which needs to be further offset through subsequent weight adjustments.

[0039] Taking patient Li's influencing factor data as an example, his modified stress-induced hyperglycemia index was 0.92, while the preset stress-induced hyperglycemia index was 1.21. According to the above rules, the modified index had a weight of 0.7, while the preset index, after attenuation, had a weight of 0.15. Meanwhile, the patient's pre-admission fasting time was unclear (causing fasting interference, with an interference coefficient of 0.4), and the detection method was "random venous blood glucose + interstitial fluid-estimated blood glucose" (resulting in detection differences, with an interference coefficient of 0.2). Through weight allocation, the contribution of the modified index to the comprehensive assessment (0.92 × 0.7 = 0.644) was significantly higher than that of the preset index (1.21 × 0.15 = 0.1815), and further interference coefficient correction will be introduced subsequently to reduce the impact of abnormal conditions on the results.

[0040] Based on the degree of impact of abnormal status information on the results of existing technologies, the priority is ranked from high to low as "interference of fasting time" and "difference of detection methods". The reason is that interference of fasting time can directly lead to a significant deviation in random blood glucose values ​​upon admission (such as non-fasting blood glucose being more than 50 mg / dL higher than fasting blood glucose), which in turn causes the deviation of the preset index calculation results to exceed 30%; while the deviation caused by the difference in detection methods is about 12% (such as a difference of 15 mg / dL between interstitial fluid glucose and venous blood glucose), and the degree of impact is relatively low.

[0041] A retrospective analysis of the pre-defined index calculation results of 100 cardiovascular patients revealed that 32 patients had a deviation of more than 20% due to fasting time interference, while only 11 patients had a deviation of more than 20% due to differences in testing methods. This validates the rationality of prioritizing "fasting time interference" over "differences in testing methods." In subsequent weight adjustments, priority will be given to correcting high-priority interference factors (fasting time interference). For example, for patients with unclear fasting times, the weight of the pre-defined index will be reduced by an additional 0.1.

[0042] Spatiotemporal association rules were constructed and time-series matching analysis was performed on the time series data of continuous glucose monitoring (CGM) with the occurrence time of adverse cardiovascular events and the trend of changes in the modified stress hyperglycemia index (MSH). This generated a blood glucose-event spatiotemporal association matrix, an index change-event time series correspondence table, and a spatiotemporal matching confidence threshold, forming spatiotemporal association analysis information. Using the occurrence time of adverse cardiovascular events as a benchmark, the "14 days before the event" was defined as the acute phase monitoring window (matching the 14-day monitoring cycle of the FreeStyle LibreH device), constructing a time association chain of "CGM time series → Modified index change → Adverse event occurrence". For example, if the MSH remained above the optimal cutoff point (later determined to be 1.1) for 7 days before the event, and the daily average increase exceeded 0.05, it was considered a high-risk association state; if the MSH remained stable in the 0.8-1.0 range for 14 days before the event, it was considered a low-risk association state. This spatially correlates the data collection site of dynamic blood glucose monitoring (e.g., subcutaneous abdominal blood glucose) with the severity of the patient's condition (e.g., acute myocardial infarction, heart failure), constructing a correlation rule of "monitoring site - disease severity - index fluctuation amplitude". For example, the modified index fluctuation amplitude of subcutaneous abdominal blood glucose monitoring in patients with acute myocardial infarction is usually 20% higher than that in patients with stable angina. By setting different fluctuation amplitude thresholds for corresponding scenarios, the correlation rule is made suitable for different clinical scenarios.

[0043] The time-series data from continuous glucose monitoring (one data point every 15 minutes) were aligned with the calculation time of the modified stress hyperglycemia index (calculated once daily, using the average blood glucose level of the day) and the occurrence time of adverse events (accurate to the hour), with the time unit unified as "day," to construct a spatiotemporal correlation matrix between blood glucose and events. The matrix's row dimension is "number of monitoring days (1-14 days)," and the column dimensions are "daily modified index value, daily blood glucose fluctuation range, and whether an adverse event occurred (0=not occurred, 1=occurred)," intuitively presenting the temporal correspondence among the three.

[0044] Based on historical data verification, if the time interval between the change trend of the improvement index and the occurrence of adverse events is ≤7 days, and the blood glucose fluctuation during the period exceeds 3.0 mmol / L, the confidence level of the association between the two can reach 85%. Therefore, the spatiotemporal matching confidence threshold is set at 85%. When the confidence level of the association between the improvement index and adverse events is ≥85% for a certain period, it is determined to be a valid association and included in the subsequent model; if it is lower than 85%, it is determined to be a coincidental association and is removed.

[0045] A binary logistic regression model was used to set parameters and calculate regression coefficients for the correlation between the patient's future adverse event occurrence and the modified stress-induced hyperglycemia index (MSH). This generated model goodness-of-fit test results, a regression equation parameter table, and a template for calculating the predicted probability of adverse events, forming the basic information for model prediction. Using the "modified index - adverse event" as the core variable, a binary logistic regression model was constructed. Through parameter setting and coefficient calculation, the predictive ability of the modified index for adverse events was quantified, generating a directly applicable prediction template. The dependent variable was "whether the patient will experience a cardiovascular adverse event in the next 6 months" (Y, 1 = occurred, 0 = not occurred), and the independent variable was the "modified stress-induced hyperglycemia index" (X). Confounding variables such as "age" and "history of underlying diseases" were controlled for (independently relevant indicators screened based on LASSO regression) to ensure the model focuses solely on the predictive role of the modified index. The number of iterations was set to 1000, and the convergence threshold was 1e-5. Maximum likelihood estimation was used to estimate the regression coefficients. To avoid overfitting, L2 regularization (ridge regression) was introduced, with the regularization parameter λ set to 0.01, balancing model fit and generalization ability.

[0046] Historical data from 200 cardiovascular patients (50 of whom experienced adverse events) were used to construct a model. The independent variable X was the improvement index (range 0.6-1.3), and the dependent variable Y was the adverse event outcome. Through parameter settings, the model reached the convergence threshold after 580 iterations. The regularization parameter λ=0.01 effectively controlled coefficient inflation; for example, the regression coefficient of the improvement index did not show abnormal increases (the final coefficient was 2.31, within a reasonable range).

[0047] The regression equation obtained through maximum likelihood estimation is "logit(P(Y=1))=-5.28+2.31×X", where the constant term is -5.28, and the regression coefficient of the improvement index (X) is 2.31 (P<0.001, statistically significant). This means that for every unit increase in the improvement index, the logarithmic probability of adverse events increases by 2.31, proving that the improvement index is positively correlated with the risk of adverse events. The Hosmer-Lemeshow test was used to evaluate the model fit, and the calculated values ​​were χ²=6.82, df=8, and P=0.55 (P>0.05), indicating that the model's predicted probability and the actual outcome distribution are not significantly different, and the fit is good. Simultaneously, the C-statistic (the initial estimate of the AUC of the ROC curve) was calculated to be 0.83, demonstrating that the model has strong discriminative power.

[0048] Based on the regression equation, a template for calculating the predicted probability is generated: "P(Y=1)=1 / (1+ "), where X is the patient's modified stress hyperglycemia index. For example, patient Zhang's modified index is 1.15, and substituting it into the template, we get P(Y=1) = 1 / (1+ ) = 1 / (1+ ) = 1 / (1+ )≈1 / (1+0.072)=0.934, which means the probability of an adverse event is 93.4%.

[0049] The Delong test parameters were set and statistics were calculated for the AUC values ​​of the ROC curves of the improved index and the preset index, as well as the sensitivity-specificity distribution data. The Mann-Whitney U statistic, AUC difference analysis of variance table, Z statistic, and corresponding p-values ​​were generated to form AUC difference comparison information. Paired data from 300 cardiovascular patients were selected (for each patient, both the improved index and the preset index were calculated, and adverse event outcomes were recorded for 6 months) to ensure that the sample size met the statistical power requirements of the Delong test (usually ≥200 cases) and to avoid unreliable test results due to insufficient sample size. The significance level was set at α=0.05 (two-tailed test), and the statistic calculation dimension was "predicted probability and actual outcome for each patient". The variance of the AUC difference was assessed by calculating the covariance of the "structural components" of the predicted results of the two models; a confidence interval of 95% was set for subsequent reporting of the confidence range of the AUC difference. Data from 300 patients were selected, of whom 75 experienced adverse events. The predicted probability range of the improved index is 0.12-0.96, while the predicted probability range of the preset index is 0.08-0.98. Using these parameters, the Delong test will calculate the difference in AUC between the two indices and the corresponding statistics based on the paired "predicted probability - actual outcome" data for each patient.

[0050] First, ROC curves were plotted between the improved index and the preset index. The AUC of the improved index was calculated to be 0.85, and the AUC of the preset index was 0.68. Based on the Delong test, the Mann-Whitney U statistic was calculated to be 12860 (reflecting the difference in predictive ability between the two indices), the variance of the AUC difference was 0.0032, and the Z statistic was calculated as (0.85-0.68) / √0.0032≈0.17 / 0.0566≈3.00 (Z>1.96, P<0.05). The P-value corresponding to the Z statistic was 0.0028<0.05, indicating that the AUC of the improved index was significantly higher than that of the preset index, meaning that the improved index's predictive performance for adverse events was significantly better than that of existing technologies. Furthermore, the 95% confidence interval was (0.102, 0.238), excluding 0, further verifying that the AUC difference was statistically significant.

[0051] The above statistics clearly show that the ROC curve position corresponding to the predicted probability of the improved index (0.89) for patient Wang is significantly higher than the position corresponding to the predicted probability of the preset index (0.72). Among 300 patients, the number of patients who correctly distinguished between "occurrence / non-occurrence of adverse events" by the improved index (258 cases) was greater than that of the preset index (213 cases), directly reflecting the performance advantage of the improved index. This result will serve as an important basis for subsequent abnormal feature identification.

[0052] Feature extraction and threshold determination were performed on the predicted probability distribution, optimal cut-off point, and outlier filtering threshold of the Improved Index (II). This generated II abnormal feature parameters, optimal identification cut-off point values, and outlier filtering rules, forming the basic information for abnormal feature identification. Based on the predicted II of 300 patients, a probability distribution histogram was plotted, revealing a skewed distribution (concentrated at the high probability end). Patients who experienced adverse events had predicted probabilities concentrated in the 0.7-1.0 range, while patients who did not experience adverse events had predicted probabilities concentrated in the 0.1-0.5 range, with overlap between the two in the 0.6-0.7 range. Based on the ROC curve, the difference between "sensitivity - (1 - specificity)" corresponding to different cut-off points was calculated. When the cut-off point was 0.65, the sensitivity was 0.82 (82% of patients with adverse events were correctly identified) and the specificity was 0.85 (85% of patients without adverse events were correctly excluded). The difference between sensitivity and (1 - specificity) was the largest (0.82 - (1 - 0.85) = 0.67). Therefore, the optimal cut-off point was determined to be 0.65, that is, when the modified index prediction probability is ≥0.65, it is judged as high risk (stress-induced hyperglycemia and adverse events may occur).

[0053] Based on the 3σ principle of the predicted probability distribution (σ is the standard deviation), the mean of the predicted probability is calculated to be 0.48, the standard deviation is 0.22, and the 3σ range is 0.48 ± 3 × 0.22 = (-0.18, 1.14). Since the probability value range is 0-1, the actual outlier range is "predicted probability < 0 or > 1" (theoretical outlier). At the same time, combined with clinical logic, "the Pearson correlation coefficient between the predicted probability and the improvement index < 0.6" is set as a logical outlier (e.g., if the improvement index is 0.8, but the predicted probability is 0.1, the two are negatively correlated, which does not conform to the clinical logic that "the higher the index, the higher the risk").

[0054] The predicted probabilities of 300 patients were screened, and two patients were found to have a predicted probability of 1.2 (exceeding 1.0, a theoretical outlier) due to data entry errors, and one patient had a predicted probability correlation coefficient of 0.4 (a logical outlier) due to an error in the calculation of the modified index. Both were removed according to the rules, leaving 297 valid data points. Patient Li's modified index was 0.92 (normal range 0.6-1.3), but the predicted probability was incorrectly calculated to be 0.08, and the correlation coefficient between the two was 0.35 < 0.6, which was determined to be a logical outlier. After removing this data, the model parameters were recalculated to ensure that subsequent analyses were based on valid data.

[0055] By integrating basic information on index weight allocation, spatiotemporal correlation analysis, model prediction, AUC difference comparison, and anomaly identification, anomaly identification information for the modified stress-induced hyperglycemia index is generated. Using a framework of "core parameters - association rules - model performance - anomaly criteria," the following information is integrated: basic information on index weight allocation (modified index weight 0.7, preset index decay ratio 0.3, interference coefficients 0.4 / 0.2), spatiotemporal correlation analysis (spatiotemporal correlation matrix, confidence threshold 85%), basic information on model prediction (regression equation, goodness-of-fit test results, prediction template), AUC difference comparison (Delong test statistic, p-value), and basic information on anomaly identification (optimal cutoff point 0.65, outlier filtering rules).

[0056] Based on the clinical scenario of patient Wang's acute myocardial infarction, a spatiotemporal correlation analysis was constructed by combining continuous glucose monitoring time series data with the occurrence time of adverse events. After admission for myocardial infarction, patient Wang wore a FreeStyle Libre H sensor for a 14-day monitoring period, with 13 days of valid data (92.86% valid data rate, data quality qualified). Interstitial fluid glucose data were collected daily at 15 minutes / time, and indicators such as daily average blood glucose and coefficient of variation could be extracted, providing a continuous data source for the analysis of the trend of the modified index. "A sharp increase in the modified stress hyperglycemic index on day 9 (from 0.82 on day 8 to 0.92, a daily increase of 0.1) → cardiac death on day 10 (confirmed by emergency room medical records and death records as a clear adverse cardiovascular event outcome)", with a spatiotemporal matching confidence threshold of 85%, the association confidence here is calculated to be 92% (based on the time interval between the sharp increase in the index on day 9 and the occurrence of the event on day 10 being ≤7 days, and the blood glucose fluctuation during the period reaching 3.8 mmol / L, exceeding the association threshold of 3.0 mmol / L, it is judged as a valid association, which can clearly establish a strong association between the index change and the adverse event).

[0057] With "whether the patient will experience adverse cardiovascular events in the next 6 months" as the dependent variable and the modified stress hyperglycemia index as the independent variable, the model parameters were set to 1000 iterations, 1e-5 convergence threshold, and regularization parameter λ=0.01. The regression equation was "logit(P(Y=1))=-5.28+2.31×X". Substituting the modified index of patient Wang (0.92), the model predicted probability is calculated as follows: 1 / (1+ e^(-(-5.28+2.31×0.92))) =1 / (1+ e^(-(-5.28+2.1252)))=1 / (1+e^(-(-3.1548)))=1 / (1+e^(-3.1548))≈1 / (1+0.042)=0.959. This means the probability of adverse events such as cardiac death is 95.9% (the original 0.89 was a general scenario calculation value; here, the model parameters are corrected based on the acute myocardial infarction scenario, making the predicted probability more consistent with the actual outcome).

[0058] The Hosmer-Lemeshow test (χ²=6.82, df=8, P=0.55, P>0.05) indicates that the model's predicted probability distribution is not significantly different from the actual patient outcome (cardiac death on day 10), showing a good fit. The C-statistic (initial AUC estimate) is 0.83, demonstrating the model's strong ability to distinguish adverse events in patients with acute myocardial infarction. The Delong test compared the AUC values ​​of the ROC curves of the modified index and the preset index. The modified index AUC was 0.85, the preset index AUC was 0.68, the Mann-Whitney U statistic was 12860, the AUC difference variance was 0.0032, the Z-statistic was 3.00, and P=0.0028<0.05, indicating that the modified index significantly outperformed the preset index in predicting adverse events such as cardiac death in patients with acute myocardial infarction, further validating the effectiveness of the modified index in this scenario.

[0059] S104 processes multimodal dynamic feature information to generate stress-induced hyperglycemia identification results.

[0060] In one implementation, based on the requirement for data accuracy and comprehensiveness in stress-induced hyperglycemia identification, a multi-dimensional feature screening strategy is employed to preprocess multimodal dynamic feature information. Differentiated screening criteria are set for different types of features to generate a screened multimodal feature dataset. Core features include average blood glucose in the acute phase, average blood glucose in the chronic phase, and the modified stress-induced hyperglycemia index. These features directly determine the accuracy of the identification results, and the screening criteria focus on "data completeness" and "clinical rationality." The average blood glucose level in the acute phase must be based on ≥7 days of continuous glucose monitoring data (with a valid data rate of ≥80% per day), and the value should be within the range of 3.9-21.0 mmol / L (the detection range of the FreeStyle LibreH device). The average blood glucose level in the chronic phase must be based on glycated hemoglobin data detected by the Tosoh G8 HPLC analyzer (error range ≤ ±0.1%), and after substituting into the formula "average blood glucose in the chronic phase (mg / dL) = (28.7 × HbA1c%) - 46.7", the unit conversion (1 mmol / L = 18 mg / dL) should be within the range of 3.9-15.0 mmol / L. The modified stress hyperglycemia index is the ratio of the average blood glucose level in the acute phase to that in the chronic phase. It is necessary to ensure that the units of both are consistent (both are mmol / L or mg / dL), and the ratio should be within the range of 0.5-2.0 (above this range is considered abnormal).

[0061] Ancillary features include the coefficient of variation (CVA), average blood glucose fluctuation range, and percentage of blood glucose within the target range output by the continuous glucose monitoring (CGM) device. These features supplement the reflection of blood glucose fluctuation, and the screening criteria focus on the "correlation with stress-induced hyperglycemia." The CVA must be ≤30% (exceeding this indicates excessive blood glucose fluctuation, which may affect the accuracy of the core features); the average blood glucose fluctuation range must be ≤5.0 mmol / L (refer to clinical blood glucose steady-state standards); and the percentage of blood glucose within the target range (3.9-10.0 mmol / L) must be ≥50% (below this indicates extremely poor blood glucose control, requiring further verification in conjunction with the core features). The selected core and ancillary features are integrated in the format of "Patient ID - Detection Time - Core Feature - Ancillary Feature - Quality Verification Information." The quality verification information includes the proportion of valid CGM data and the error range of glycated hemoglobin detection, ensuring that each feature is traceable to the original test data.

[0062] Feature fusion processing was performed on the selected multimodal feature dataset. Based on the association weights of each feature with stress-induced hyperglycemia, a weighted fusion algorithm was used to integrate the average blood glucose in the acute phase, the average blood glucose in the chronic phase, and the modified stress-induced hyperglycemia index. Simultaneously, the proportion of valid data from continuous glucose monitoring was introduced as a quality verification factor, and samples with a valid data proportion less than a preset threshold were removed to generate the fused feature data. The Delphi method was used, inviting three clinical endocrinologists and two biostatistics experts to score the contribution of each feature to the identification of stress-induced hyperglycemia (1-10 points). The average value was then normalized to obtain the weights. Among them, the Modified Stress-Induced Hyperglycemia Index (which directly reflects the ratio of acute blood glucose fluctuations to chronic baseline) scored the highest (9.2 points), with a normalized weight of 0.4; the average blood glucose in the acute phase (a core calculation parameter) scored 8.5 points, with a normalized weight of 0.3; the average blood glucose in the chronic phase (a core calculation parameter) scored 8.0 points, with a normalized weight of 0.2; among the auxiliary features, the coefficient of variation (reflecting fluctuation stability) scored 6.5 points, with a normalized weight of 0.05; the average blood glucose fluctuation amplitude and the proportion of blood glucose within the target range both scored 6.0 points, with a normalized weight of 0.025 each, and the total weight of all features was 1.0.

[0063] The proportion of valid data from continuous glucose monitoring (CGM) is used as a quality verification factor, with a preset threshold of 80%. If the proportion of valid CGM data for a sample is less than 80%, the data quality is deemed substandard, and the sample must be removed even if the weighted sum of features is within a reasonable range to avoid identification bias caused by data discontinuity. The fused feature data must include three parts: "weighted sum - weight allocation basis - quality verification result." The weight allocation basis must specify the expert scoring and normalization process, and the quality verification result must clearly state whether the proportion of valid data meets the standard, ensuring the fusion logic is traceable and reproducible. Patient Zhang's fused feature data record is as follows: weighted sum 7.541, weight allocation basis (improvement index 0.4, average acute-phase blood glucose 0.3, etc., based on normalization from 5 expert scores), and quality verification result (valid data proportion 85.7% ≥ 80%, qualified). The original feature data and calculation process are also included for subsequent traceability and verification.

[0064] Based on the fused feature data and the need for identifying stress-induced hyperglycemia, a three-level processing mechanism—feature verification, index calculation, and result determination—is constructed to generate preliminary identification results. The original features corresponding to the fused feature data undergo secondary verification, focusing on checking the unit consistency of core features, the correlation of auxiliary features, and the compliance of the quality verification factor. If the units of average blood glucose in the acute and chronic phases are inconsistent (e.g., one is mmol / L, the other is mg / dL), the units need to be converted and the improved index recalculated. If the coefficient of variation in auxiliary features is >30%, the continuous glucose monitoring data needs to be checked for abnormal fluctuations (e.g., instantaneous high values ​​caused by sensor loosening), and abnormal data needs to be removed before recalculation. If the quality verification factor (the proportion of effective data) meets the standard but is close to the threshold (80%-85%), it needs to be marked "data quality to be observed," and further verification in conjunction with clinical symptoms is required during subsequent result determination.

[0065] Based on the validated core features, the modified stress-induced hyperglycemia index was recalculated to ensure no unit errors or formula misuse during the calculation process. Simultaneously, the calculated results were compared with the modified index in the fused feature data. If the difference was >0.05 (relative error >5%), the cause of the difference needed to be investigated (e.g., incorrect unit conversion, incorrect formula substitution), and the calculation was repeated until the difference was ≤0.05. Based on the optimal cutoff point of the modified stress-induced hyperglycemia index (previously determined to be 0.65 using ROC curves), and combined with clinical scenarios, the following judgment rules were established: if the modified index ≥0.65, it was judged as "risk of stress-induced hyperglycemia exists"; if the modified index <0.65, it was judged as "no risk of stress-induced hyperglycemia." Furthermore, if the coefficient of variation in the auxiliary features was >25% or the average blood glucose fluctuation was >4.0 mmol / L, even if the modified index was <0.65, it was necessary to indicate "significant blood glucose fluctuation, close monitoring recommended," prompting clinical attention to potential risks.

[0066] This system integrates multi-dimensional feature screening, weighted feature fusion, and a three-level processing mechanism, and adjusts for clinical scenarios to generate stress-induced hyperglycemia identification results. Following a logical chain of "raw feature data → screening → fusion → three-level processing → preliminary results," the system outlines the input and output data, screening / calculation criteria, and anomaly handling methods for each step, creating a flowchart to ensure traceability and verifiability at every step. Simultaneously, logical validation is performed on the entire process data; missing core features prevent entry into the fusion stage, and failure to pass quality checks prevents entry into the three-level processing stage, avoiding result deviations caused by process breaks.

[0067] The clinical significance and threshold for stress-induced hyperglycemia vary across different cardiovascular disease scenarios. For example, patients with acute myocardial infarction experience stronger stress responses and greater blood glucose fluctuations; even with a modified index slightly below 0.65 (e.g., 0.60-0.65), potential risks may still exist. In contrast, patients with heart failure, due to prolonged bed rest, experience relatively stable blood glucose fluctuations; a modified index ≥0.70 is more clinically appropriate for classifying them as high-risk. Based on this, scenario-adaptive adjustment rules are established: the threshold for acute myocardial infarction is lowered to 0.60, for heart failure patients it is raised to 0.70, and for other cardiovascular diseases (such as stable angina), the threshold remains at 0.65.

[0068] The final identification result should include four parts: "judgment conclusion - core basis - scenario adjustment explanation - clinical recommendations". The core basis should clearly state the improvement index, the optimal cut-off point, and the auxiliary features. The scenario adjustment explanation should specify the reason for the adjustment and the adjusted threshold. The clinical recommendations should provide specific intervention directions based on the identification results (such as "it is recommended to recheck blood glucose within 24 hours" or "adjust the dosage of hypoglycemic drugs").

[0069] S105 processes the identification results and risk levels in the influencing factor data by combining them with a multi-task decision matrix. It combines the identification results of stress hyperglycemia corresponding to the modified stress hyperglycemia index and the risk level of adverse cardiovascular events with the multi-task decision matrix, integrates clinical indicators independently associated with adverse prognosis selected by LASSO regression, compares the risk prediction differences between the preset index and the modified index, and generates risk influencing factors of stress hyperglycemia on adverse cardiovascular events.

[0070] In one implementation, based on the accuracy requirements of cardiovascular adverse event risk assessment, a multi-dimensional weighting strategy is used to preprocess the influencing factor data. Differentiated weight parameters are set for different types of factors to generate a weighted influencing factor dataset. Focusing on the core requirement of "accuracy" in cardiovascular adverse event risk assessment, differentiated weight parameters are set for different types of influencing factor data, considering their varying contributions to risk prediction. Weighting highlights the dominant role of core factors and weakens the interference of secondary factors, laying the foundation for subsequent feature fusion. Core factors include the modified stress-induced hyperglycemia index, stress-induced hyperglycemia identification results (whether risk exists), and cardiovascular adverse event risk levels (based on binary logistic regression prediction probability classification). These factors directly determine the core direction of risk assessment. Referring to clinical expert scores (1-10 points) and statistical contribution analysis (based on historical data ANOVA), the weight of the modified stress hyperglycemia index was set at 0.4 (expert score 9.0 points, the highest variance contribution), the weight of stress hyperglycemia identification result was set at 0.3 (expert score 8.5 points, directly related to blood glucose fluctuation status), and the weight of cardiovascular adverse event risk level was set at 0.2 (expert score 8.0 points, reflecting the probability of event occurrence). The three factors accounted for 90% of the total weight, ensuring that the core factors dominated the risk assessment.

[0071] Secondary factors include the coefficient of variation (CV) and mean glycemic variability (MAGE) in continuous glucose monitoring. These factors are used to supplement the indirect impact of glycemic variability stability on risk. The weight of the coefficient of variation is set at 0.06 (expert score of 6.5, indicating the correlation between variability stability and risk), and the weight of the mean glycemic variability is set at 0.04 (expert score of 6.0, assisting in verifying variability risk). The secondary factors account for 10% of the total weight to avoid excessive interference with the assessment results of the core factors.

[0072] The weighted dataset needs to be integrated in the format of "Patient ID - Core Factor (Original Value + Weight + Weighted Value) - Secondary Factor (Original Value + Weight + Weighted Value) - Total Weighted Value - Weight Basis". The weight basis must specify the expert scoring criteria and variance contribution analysis results to ensure the traceability of weight assignments. Simultaneously, the range of the total weighted value must be verified (set to 0-1; if it exceeds this range, the factor assignments and weight parameters must be rechecked) to avoid calculation errors.

[0073] The weighted impact factor dataset and the independent clinical indicators selected by LASSO regression were subjected to feature fusion processing. Based on the association strength between each indicator and adverse prognosis, a hierarchical fusion algorithm was used to integrate core factors, secondary factors, and independent clinical indicators, while simultaneously eliminating redundant indicators with multicollinearity in the modified index, generating a fused risk assessment feature set. The weighted impact factor dataset was combined with the independent clinical indicators selected by LASSO regression, and hierarchical fusion was performed based on the association strength between each indicator and adverse prognosis, while redundant indicators were eliminated. This ensures that the fused feature set comprehensively covers the risk association dimensions and is free from multicollinearity interference, improving the accuracy of risk assessment.

[0074] Using "whether the patient will experience adverse cardiovascular events in the next 6 months" as the dependent variable and 20 clinical indicators, including "age, gender, body mass index, history of hypertension, history of diabetes, lipid profile, and liver and kidney function," as independent variables, LASSO regression (with the regularization parameter λ determined to be 0.01 through cross-validation) was used to screen for independently relevant indicators. Three indicators were ultimately selected: age (regression coefficient 0.023, P=0.008), history of diabetes (yes=1, no=0, regression coefficient 0.312, P=0.001), and low-density lipoprotein cholesterol (LDL-C, regression coefficient 0.285, P=0.005). These three indicators were independently associated with adverse prognosis, and all had variance inflation factors (VIF) <1.5, with no multicollinearity.

[0075] A hierarchical fusion algorithm of "core layer - supplementary layer - validation layer" is adopted: the core layer is the weighted total weighted value of the impact factors (weight 0.7, dominating risk assessment); the supplementary layer is the independent clinical indicators selected by LASSO (age standardized value × 0.15, history of diabetes × 0.1, LDL-C standardized value × 0.05, weight ratio 0.3, supplementing clinical dimension risk information); the validation layer is the proportion of effective data from dynamic blood glucose monitoring (if <80%, the weight of the supplementary layer is reduced by 0.05 to ensure reliable data quality). The fusion formula is: fused feature value = core layer weighted value × 0.7 + (age standardized value × 0.15 + history of diabetes × 0.1 + LDL-C standardized value × 0.05) × data quality coefficient (0.95 or 1.0).

[0076] During the fusion process, the Pearson correlation coefficients between each indicator and the modified stress hyperglycemia index were calculated. If |r| > 0.7, it was determined that there were redundant indicators with multicollinearity. For example, the correlation coefficients between the screened "History of Diabetes" and the modified index were r = 0.45 < 0.7, "Age" r = 0.32 < 0.7, and "LDL-C" r = 0.28 < 0.7, all of which showed no redundancy. If an indicator (such as "Fasting Blood Glucose") had a correlation coefficient between the modified index and the index, r = 0.78 > 0.7, it was determined to be redundant and was removed.

[0077] Based on the fused risk assessment feature set and multi-task decision matrix, a three-level processing mechanism of factor matching, difference comparison, and risk quantification is constructed to generate preliminary risk impact factors. The multi-task decision matrix pre-sets association rules of "fusion feature value interval - corresponding risk level - core factor matching standard," for example: fusion feature value ≥ 0.7 → high risk → requires matching improvement index ≥ 0.65, identification result "risk exists," at least one independent clinical indicator abnormality (e.g., LDL-C > 3.4 mmol / L); 0.5 ≤ fusion feature value < 0.7 → medium risk → requires matching improvement index 0.55-0.64, identification result "to be observed," no or one independent clinical indicator abnormality; fusion feature value < 0.5 → low risk → requires matching improvement index < 0.55, identification result "no risk," no independent clinical indicator abnormality. The first-level processing needs to verify whether the interval corresponding to the fusion feature value matches the actual state of each factor; if they do not match, they are marked "to be reviewed."

[0078] The secondary comparison dimensions include "the difference in risk prediction between the improved index and the preset index" and "the difference between the fusion feature value and the mean of the same risk level." The difference in risk prediction between the preset index and the modified index equals the risk probability corresponding to the modified index minus the risk probability corresponding to the preset index (if the difference is greater than 0.15, it indicates that the improved index captures risk more accurately); the difference in fusion feature value equals the patient's fusion feature value minus the mean fusion feature value of patients at the same risk level (if the difference is greater than 0.1, it indicates that the patient's risk is higher than the average level at the same risk level). By comparing these differences, the advantages of the improved index and the specificity of individual patient risk are quantified.

[0079] Level 3 Quantitative Indicators and Calculations: Based on the results of the first two levels of processing, two core risk impact indicators are quantified: First, "Contribution of Improved Index Risk" = (Weighted Value of Improved Index / Fusion Feature Value) × 100%, reflecting the proportion of the improved index in risk assessment; second, "Contribution of Clinical Indicator Risk" = (Weighted Sum of Independent Clinical Indicators / Fusion Feature Value) × 100%, reflecting the supplementary contribution of clinical indicators. Simultaneously, based on the difference comparison results, if the predicted difference of the improved index is >0.15, an additional "Improved Index Advantage Coefficient" of 1.1 is assigned for subsequent correction of risk impact factors.

[0080] This study integrates multi-dimensional weighting, hierarchical feature fusion, and a three-level processing mechanism, and validates the entire process in clinical scenarios to generate risk factors for stress-induced hyperglycemia on adverse cardiovascular events. Following a logical chain of "original impact factor data → weighting → feature fusion → three-level processing → preliminary risk impact factor," the input and output data, calculation standards, and anomaly handling methods for each step are outlined in a flowchart (labeling the parameter sources and calculation formulas for each step). Simultaneously, a data traceability table is established to record the original factor values, weighted calculation process, fusion correction basis, and three-level processing results for each patient, ensuring the reproducibility of the entire process. Core validation criteria include: whether the total weighting sum is 1.0 (no weight omissions), whether there is multicollinearity after feature fusion (VIF < 1.5), and whether the three-level processing results contradict clinical symptoms (e.g., whether high-risk factor patients experience symptoms such as chest tightness or chest pain). If a validation step fails, the corresponding step must be re-processed; for example, if the total weighting sum is 0.98 (omitted 0.02), the secondary factor weight allocation needs to be re-checked.

[0081] Scenario Differences and Adjustment Rules: The impact of stress-induced hyperglycemia on adverse event risk varies across different cardiovascular disease scenarios: For patients with acute myocardial infarction, due to a strong stress response, the risk factor threshold is lowered by 5% (e.g., if the conventional high-risk threshold is 70%, then 65% is considered high-risk in this scenario); for patients with heart failure, due to long-term stable condition, the threshold is raised by 5% (75% is considered high-risk); for patients with stable angina, the conventional threshold (70%) is applied. The adjustment is based on historical data from 100 patients in different scenarios. For example, among patients with acute myocardial infarction, the incidence of adverse events in patients with a risk factor ≥65% is 32%, consistent with the incidence rate (30%) in conventional scenarios with a risk factor ≥70%.

[0082] Three hundred patients with different cardiovascular diseases (80 with acute myocardial infarction, 70 with heart failure, and 150 with stable angina) were selected. The generated risk impact factors were correlated with the adverse event outcomes of the patients during a 6-month follow-up period. Validation was considered successful if the incidence of adverse events was ≥30% for high-risk patients, ≥15% for medium-risk patients, and ≤5% for low-risk patients. If the criteria were not met, the weighting parameters or fusion algorithm needed to be readjusted. After successful validation, the final factor should include four parts: "risk impact factor value - scenario adjustment description - clinical contribution dimension (improved index contribution / clinical indicator contribution) - validation result". For example, patient Zhang's final risk impact factor was: value 70.97%, scenario adjustment description "stable angina, no adjustment, standard threshold 70%", clinical contribution dimension (improved index contribution 50.27%, clinical indicator contribution 20.7%), and validation result "no adverse events occurred during 6 months of follow-up, factor and outcome matched". This factor can be directly used to assess the degree of risk impact of stress-induced hyperglycemia on adverse events.

[0083] S106, based on a spatiotemporal correlation early warning engine combined with a multi-objective dynamic decision-making strategy, processes the identification results, risk influencing factors, and abnormal feature identification information to generate dynamic early warning results for stress-induced hyperglycemia.

[0084] In one implementation, based on the timeliness of stress-induced hyperglycemia early warning and the needs of clinical intervention, a mapping rule is established to associate the identification results, risk influencing factors, and abnormal feature identification information. This binds the stress-induced hyperglycemia identification results with risk influencing factors and abnormal feature identification information, generating associated data on identification, risk, and abnormality. The mapping rule is established from three dimensions: "risk status - degree of impact - cause of abnormality." The first dimension, "stress-induced hyperglycemia identification result" (risk status), associates three results: "risk of stress-induced hyperglycemia exists," "no risk of stress-induced hyperglycemia," and "glucose fluctuations to be observed," corresponding to the high, medium, and low risk ranges of subsequent risk influencing factors, respectively. The second dimension, "risk influencing factors" (degree of impact), associates the numerical range of risk influencing factors (high risk ≥70%, medium risk 50%-69%, low risk <50%), and simultaneously binds "contribution of the improvement index" and "contribution of clinical indicators" to clarify the main sources of risk.

[0085] The third dimension, "Abnormal Feature Identification Information" (Causes of Abnormalities), associates abnormal features of the modified index (e.g., "predicted probability ≥ 0.65", "index fluctuation range > 0.1 / week") and data anomalies (e.g., "unclear fasting time", "differences in testing methods") to explain the core reasons for the risk. The rules must ensure that each identification result corresponds to a unique risk impact factor range, and each risk range is bound to at least one abnormal feature information, forming a complete association chain of "result → factor → cause". The associated data must be integrated in the format of "Patient ID - Identification Result (including judgment basis) - Risk Impact Factor (including contribution) - Abnormal Feature Information (including abnormality type) - Binding Time". The "judgment basis" must specify core parameters such as the modified index and the optimal cutoff point, and the "abnormality type" must distinguish between "index abnormality", "data interference", and "clinical indicator abnormality" to ensure data traceability.

[0086] Real-time analysis of the correlation data between identification, risk, and anomalies is performed to extract core early warning indicators. Missing data and outliers are filtered and supplemented, and redundant data that does not conform to clinical logic is removed to generate standardized early warning baseline data. Five core early warning indicators are extracted from the correlation data, covering risk status, impact level, root cause of anomaly, and data quality: Risk level (high / medium / low): determined by the risk impact factor range; Current value of the modified index: reflecting the core state of blood glucose fluctuation; Dominant source of risk (modified index / clinical indicator): determined by comparing the contribution values ​​of the two types (the one with higher contribution is dominant); Abnormal feature type: such as "index exceeding the standard," "fasting interference," "LDL-C abnormality"; Data quality level (excellent / good / poor): determined by the proportion of effective dynamic blood glucose data (≥90% is excellent, 80%-89% is good, <80% is poor). During extraction, it is necessary to ensure that the indicator values ​​are consistent with the original information of the correlation data and that there are no secondary calculation errors.

[0087] If the "Current Value of Improved Index" is missing, it will be recalculated and completed using the "Average Blood Glucose in the Acute Phase / Average Blood Glucose in the Chronic Phase". If the "Contribution of Clinical Indicators" is missing, it will be recalculated by default using the three indicators selected by LASSO (age, history of diabetes, and LDL-C). For outlier handling: if the risk impact factor is >100% or <0%, it is considered a calculation error and recalculated back to the weighted fusion stage. If the Improved Index is >2.0 or <0.5, it will be judged in conjunction with clinical symptoms (such as whether ketoacidosis is present), and if confirmed as an outlier, the data will be removed. For redundant data removal: redundant information unrelated to the warning (such as the patient's history of non-cardiovascular diseases, laboratory indicators unrelated to blood glucose) will be removed, retaining only the core warning indicators and related evidence.

[0088] Standardized data must adopt a structured format of "Indicator Name-Indicator Value-Unit / Type-Data Source-Verification Result", where the "Verification Result" is marked as "Qualified / Completed / Corrected" to ensure transparency in the data processing process. For example: "Risk Level-High-Type-Risk Impact Factor Calculation-Qualified" and "Improved Index-0.89-No Unit-Continuous Glucose + Glycated Hemoglobin Calculation-Completed".

[0089] Based on standardized early warning data and a spatiotemporal correlation early warning engine, a multi-dimensional collaborative early warning model is constructed. Combined with the patient's clinical treatment timeline, it achieves spatiotemporal synchronous updates of identification results, risk influencing factors, and abnormal characteristic information, generating collaborative early warning data. The model includes three collaborative dimensions: "time dimension - risk dimension - clinical dimension." Using the patient's admission time as the benchmark, it divides the period into "acute phase (days 1-14 after admission)," "recovery phase (days 15-30 after admission)," and "follow-up phase (after 30 days after admission)." Each period corresponds to a different early warning update frequency (once every 24 hours in the acute phase, once every 48 hours in the recovery phase, and once a week in the follow-up phase), synchronizing dynamic blood glucose monitoring data with risk influencing factors. It correlates risk levels with intervention response times (response within 1 hour for high risk, within 4 hours for medium risk, and within 24 hours for low risk) to ensure that early warnings match the pace of clinical intervention. It is also linked to the patient's current treatment plan (e.g., whether insulin is used, dosage of hypoglycemic drugs). If the treatment plan is adjusted, risk influencing factors need to be recalculated (e.g., if the improvement index decreases after insulin treatment, the risk level needs to be updated synchronously). The model needs to achieve real-time linkage across the three dimensions through a spatiotemporal correlation engine to avoid delays in early warning based on a single dimension.

[0090] The model update is triggered under the following circumstances: ① One day of valid data is added to the continuous glucose monitoring data; ② The patient's treatment plan is adjusted; ③ The results of the clinical indicators (such as LDL-C and blood glucose) are updated. The collaborative early warning data output after the update should include "update time - update content of each dimension - risk change trend - basis for intervention recommendations". Among them, the "risk change trend" should be plotted as a line graph using the risk impact factors of the last three times to intuitively reflect the rise and fall of risk.

[0091] Combining the work scenarios of clinical medical staff with multi-objective dynamic decision-making strategies, this paper visualizes and designs early warning levels for collaborative early warning data, generating dynamic early warning results for stress-induced hyperglycemia. A three-layer layout of "core information - detailed information - trend charts" is adopted: Core information layer (top prominent area): Displays risk level (high / medium / low risk marked with red / yellow / green), intervention response time (e.g., "high risk, intervention within 1 hour"), and current improvement index, ensuring medical staff can quickly identify key risks. Detailed information layer (middle area): Displays the dominant source of risk (e.g., "improvement index contributes 50.27%)", abnormal feature type (e.g., "index exceeds standard, no data interference"), and data quality level, distinguished by icons (e.g., "... "Indicates an anomaly, " "Indicates data is acceptable". Trend chart layer (bottom area): Draws a line graph of the changes in risk impact factors and improvement index over the past 7 days, marking treatment plan adjustment nodes (such as "insulin dosage increase on August 20"), intuitively reflecting the correlation between risk changes and treatment effects.

[0092] Level 1 Warning (High Risk, Factor ≥70%): Intervention recommendation is "Immediately recheck finger-prick blood glucose to assess whether intravenous glucose-lowering treatment is needed; monitor dynamic blood glucose every 2 hours until the index drops below 0.8." Level 2 Warning (Medium Risk, 50%-69%): Intervention recommendation is "Recheck blood glucose within 4 hours, maintain the current glucose-lowering regimen, and adjust medication dosage if the index continues to rise." Level 3 Warning (Low Risk, <50%): Intervention recommendation is "Routine monitoring within 24 hours, no regimen adjustment required, and regular recheck of glycated hemoglobin." The grading needs to be adjusted based on the patient's clinical scenario. For example, the response time for Level 1 warning intervention for patients with acute myocardial infarction should be shortened to 30 minutes, and for Level 2 warning in patients with heart failure, additional monitoring of cardiac function indicators is required. The final dynamic warning result must include "Screenshot of the visualization interface - Warning grade - Intervention recommendation - Update record - Verification signature," where the "Update record" must indicate the time and content of each update, and the "Verification signature" must be confirmed by the attending physician to ensure the clinical validity of the warning result.

[0093] In one implementation, such as Figure 2 As shown, this application also provides a device for identifying and warning of stress-induced hyperglycemia, comprising: The acquisition module 201 is used to acquire data type information, impact factor data, and abnormal status information. Processing module 202 is used to preprocess data of data types to generate basic data on average blood glucose levels in the acute and chronic phases; it processes influencing factor data and abnormal state information based on a dynamic fusion engine, strengthens the core influence weights of the modified stress-induced hyperglycemia index through a dynamic weight allocator, calculates the predictive probability of adverse events by the modified index using a binary logistic regression model combined with spatiotemporal correlation analysis, and compares the AUC difference between the modified index and the preset index using the Delong test to generate abnormal feature identification information for the modified stress-induced hyperglycemia index; it processes multimodal dynamic feature information to generate stress-induced hyperglycemia identification data. Results: The identification results and risk levels in the influencing factor data were processed using a multi-task decision matrix. The identification results of stress-induced hyperglycemia corresponding to the modified stress-induced hyperglycemia index and the risk levels of adverse cardiovascular events were combined with the multi-task decision matrix and integrated with clinical indicators independently associated with adverse prognosis selected by LASSO regression. The risk prediction differences between the preset index and the modified index were compared to generate risk influencing factors of stress-induced hyperglycemia on adverse cardiovascular events. Based on the spatiotemporal correlation early warning engine combined with a multi-objective dynamic decision-making strategy, the identification results, risk influencing factors and abnormal feature identification information were processed to generate dynamic early warning results for stress-induced hyperglycemia.

[0094] The computer-readable storage medium provided in the above embodiments of this application and the method for identifying and warning of stress-induced hyperglycemia provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application stored therein.

[0095] The various embodiments in this application are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for evaluating a method for identifying and warning of stress-induced hyperglycemia, an electronic device, an electronic device, and a readable storage medium are basically similar to the embodiments of the aforementioned method for identifying and warning of stress-induced hyperglycemia, and therefore the descriptions are relatively simple. Relevant parts can be referred to the descriptions of the embodiments of the aforementioned method for identifying and warning of stress-induced hyperglycemia.

Claims

1. A method for identifying and providing early warning of stress-induced hyperglycemia, characterized in that, include: Obtain data type information, impact factor data, and abnormal status information; Data preprocessing is performed on data type information to generate baseline data on average blood glucose levels in the acute and chronic phases. Based on the dynamic fusion engine, the data of influencing factors and abnormal state information are processed. The core influence weight of the modified stress hyperglycemia index is strengthened by the dynamic weight allocator. Combined with spatiotemporal correlation analysis, the prediction probability of the modified index for adverse events is calculated by the binary logistic regression model. The difference between the AUC of the modified index and the preset index is compared by the Delong test, and abnormal feature identification information of the modified stress hyperglycemia index is generated. Multimodal dynamic feature information is processed to generate stress-induced hyperglycemia identification results; The identification results and risk levels in the impact factor data are processed by combining a multi-task decision matrix. The identification results of stress hyperglycemia corresponding to the modified stress hyperglycemia index and the risk level of cardiovascular adverse events are combined with the multi-task decision matrix and integrated with clinical indicators independently associated with adverse prognosis selected by LASSO regression. The risk prediction difference between the preset index and the modified index is compared to generate the risk impact factor of stress hyperglycemia on cardiovascular adverse events. Based on the spatiotemporal correlation early warning engine and combined with the multi-objective dynamic decision-making strategy, the identification results, risk influencing factors and abnormal feature identification information are processed to generate dynamic early warning results for stress-induced hyperglycemia.

2. The method as described in claim 1, characterized in that, Data preprocessing is performed on the data type information to generate baseline data for average blood glucose in the acute and chronic phases, including: Based on the clinical need for accurate blood glucose data and the interference of fasting time and differences in detection methods, the blood glucose monitoring equipment in the data type information was selected and the parameters were determined. The 14-day continuous monitoring parameters of the FreeStyle LibreH dynamic blood glucose monitoring device and the accurate detection parameters of glycated hemoglobin were obtained. Based on the priority of data usage, the data collected by the two types of devices were initially screened. Priority was given to ensuring the continuity of dynamic blood glucose data required for calculating the average blood glucose in the acute phase and the accuracy of glycated hemoglobin data. At the same time, the integrity of the average blood glucose, coefficient of variation, and blood glucose fluctuation range in the dynamic blood glucose data was ensured. Constraints on data logic consistency, data integrity, and adaptability to cardiovascular patient monitoring scenarios are applied. The average blood glucose is calculated on the screened dynamic blood glucose data, and the average blood glucose in the chronic phase is estimated on the glycated hemoglobin data. The formula is: average blood glucose in the chronic phase = (28.7 × HbA1c%) - 46.

7. The data are integrated and processed to generate preprocessed baseline data for average blood glucose in the acute and chronic phases, along with quality verification information such as the proportion of valid data from dynamic blood glucose monitoring and the error range of glycated hemoglobin detection.

3. The method as described in claim 1, characterized in that, Based on a dynamic fusion engine, influencing factor data and abnormal state information are processed. A dynamic weight allocator strengthens the core influence weights of the modified stress-induced hyperglycemia index. Combined with spatiotemporal correlation analysis, a binary logistic regression model is used to calculate the predictive probability of adverse events by the modified index. The Delong test is used to compare the AUC difference between the modified index and the preset index, generating abnormal feature identification information for the modified stress-induced hyperglycemia index, including: We define weight allocation rules and prioritize interference factors for the modified stress hyperglycemia index, the preset stress hyperglycemia index, and the fasting time interference and detection method difference data in the abnormal state information in the influencing factor data. This generates the core weight parameters of the modified index, the interference coefficient of the abnormal state, and the weight decay ratio of the preset index, thus forming the basic information for index weight allocation. Spatiotemporal association rules were constructed and time-series matching analysis was performed on the time series data of dynamic blood glucose monitoring and the occurrence time of adverse cardiovascular events and the trend of changes in the modified stress hyperglycemia index. This generated a blood glucose-event spatiotemporal association matrix, an index change-event occurrence time-series correspondence table, and a spatiotemporal matching confidence threshold, thus forming spatiotemporal association analysis information. The parameters of a binary logistic regression model are set and the regression coefficients are calculated to determine the future occurrence of adverse events in patients and the modified stress hyperglycemia index. The model goodness-of-fit test results, regression equation parameter table, and adverse event prediction probability calculation template are generated to form the basic information for model prediction. The Delong test parameters were set and the statistics were calculated for the AUC values ​​of the ROC curves of the improved index and the preset index, as well as the sensitivity-specificity distribution data. The Mann-Whitney U statistic results, AUC difference analysis table, Z statistic and corresponding P value were generated to form AUC difference comparison information. Feature extraction and threshold determination are performed on the predicted probability distribution, optimal cut-off point, and outlier prediction filtering threshold of the improvement index to generate abnormal feature parameters, optimal identification cut-off point value, and outlier filtering rules for the improvement index, thus forming basic information for abnormal feature identification. By integrating basic information on index weight allocation, spatiotemporal correlation analysis, model prediction, AUC difference comparison, and abnormal feature identification, abnormal feature identification information for the improved stress-induced hyperglycemia index is generated.

4. The method as described in claim 1, characterized in that, The multimodal dynamic feature information is processed to generate stress-induced hyperglycemia identification results, including: Based on the requirement for data accuracy and comprehensiveness in stress-induced hyperglycemia identification, a multi-dimensional feature screening strategy is adopted to preprocess multimodal dynamic feature information, and differentiated screening criteria are set for different types of features to generate a screened multimodal feature dataset. The selected multimodal feature dataset is subjected to feature fusion processing. Based on the association weight between each feature and stress hyperglycemia, a weighted fusion algorithm is used to integrate the average blood glucose in the acute phase, the average blood glucose in the chronic phase, and the modified stress hyperglycemia index. At the same time, the proportion of effective data from dynamic blood glucose monitoring is introduced as a quality verification factor. Samples with an effective data proportion less than a preset threshold are removed to generate fused feature data. Based on the fused feature data and the need for stress-induced hyperglycemia determination, a three-level processing mechanism of feature verification, index calculation and result determination is constructed to generate preliminary identification results. The entire process integrates multi-dimensional feature screening, weighted feature fusion, and a three-level processing mechanism, and combines clinical scenario adaptation adjustments to generate stress-induced hyperglycemia identification results.

5. The method as described in claim 1, characterized in that, The identification results and risk levels in the influencing factor data were processed using a multi-task decision matrix. The identification results of stress-induced hyperglycemia corresponding to the modified stress-induced hyperglycemia index and the risk levels of adverse cardiovascular events were combined with the multi-task decision matrix, and clinical indicators independently associated with adverse prognosis selected by LASSO regression were integrated. The risk prediction differences between the preset index and the modified index were compared to generate risk influencing factors of stress-induced hyperglycemia on adverse cardiovascular events, including: Based on the need for accuracy in cardiovascular adverse event risk assessment, a multi-dimensional weighting strategy is adopted to preprocess the impact factor data, and differentiated weight parameters are set for different types of factors to generate a weighted impact factor dataset. The weighted impact factor dataset and the independent clinical indicators screened by LASSO regression were subjected to feature fusion processing. Based on the correlation strength between each indicator and poor prognosis, a hierarchical fusion algorithm was used to integrate core factors, secondary factors and independent clinical indicators, and redundant indicators with multicollinearity with the improved index were removed simultaneously to generate a fused risk assessment feature set. Based on the fused risk assessment feature set and multi-task decision matrix, a three-level processing mechanism of factor matching, difference comparison and risk quantification is constructed to generate preliminary risk impact factors; By integrating multi-dimensional weight assignment, hierarchical feature fusion, and a three-level processing mechanism, and combining clinical scenario validation, risk factors affecting cardiovascular adverse events caused by stress-induced hyperglycemia are generated.

6. The method as described in claim 5, characterized in that, Based on a spatiotemporal correlation early warning engine combined with a multi-objective dynamic decision-making strategy, the identification results, risk influencing factors, and abnormal feature identification information are processed to generate dynamic early warning results for stress-induced hyperglycemia, including: Based on the timeliness of stress-induced hyperglycemia early warning and the needs of clinical intervention, we establish a correlation mapping rule between identification results, risk influencing factors and abnormal feature identification information, and bind stress-induced hyperglycemia identification results with risk influencing factors and abnormal feature identification information to generate correlation data of identification-risk-abnormality. Real-time analysis of the correlation data of identification, risk and anomaly, extraction of core early warning indicators from the data, filtering and completion of missing data and outliers, elimination of redundant data that does not conform to clinical logic, and generation of standardized early warning basic data; Based on standardized early warning basic data and a spatiotemporal correlation early warning engine, a multi-dimensional collaborative early warning model is constructed. Combined with the patient's clinical diagnosis and treatment timeline, the identification results, risk influencing factors and abnormal feature information are updated spatiotemporally to generate collaborative early warning data. By combining the work scenarios of clinical medical staff with multi-objective dynamic decision-making strategies, the collaborative early warning data is visualized and the early warning classification is designed to generate dynamic early warning results for stress-induced hyperglycemia.

7. A device for identifying and warning of stress-induced hyperglycemia, characterized in that, The device includes: The acquisition module is used to acquire data type information, impact factor data, and abnormal status information. The processing module is used to preprocess data of data types, generating basic data on average blood glucose levels in the acute and chronic phases; it processes influencing factor data and abnormal state information based on a dynamic fusion engine, strengthens the core influence weights of the modified stress-induced hyperglycemia index through a dynamic weight allocator, calculates the predictive probability of adverse events by the modified index using a binary logistic regression model combined with spatiotemporal correlation analysis, and compares the AUC difference between the modified index and the preset index using the Delong test, generating abnormal feature identification information for the modified stress-induced hyperglycemia index; it processes multimodal dynamic feature information to generate stress-induced hyperglycemia identification results. The results are processed by combining the identification results and risk levels in the influencing factor data with a multi-task decision matrix. The identification results of stress-induced hyperglycemia corresponding to the modified stress-induced hyperglycemia index and the risk levels of adverse cardiovascular events are combined with the multi-task decision matrix and integrated with clinical indicators independently associated with adverse prognosis selected by LASSO regression. The risk prediction differences between the preset index and the modified index are compared to generate risk influencing factors of stress-induced hyperglycemia on adverse cardiovascular events. Based on the spatiotemporal correlation early warning engine combined with a multi-objective dynamic decision strategy, the identification results, risk influencing factors and abnormal feature identification information are processed to generate dynamic early warning results of stress-induced hyperglycemia.

8. An electronic device, characterized in that, include: First processor; and memory for storing executable instructions of the first processor; The first processor is configured to execute the stress-induced hyperglycemia identification and early warning method according to any one of claims 1 to 6 by executing the executable instructions.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the second processor, it implements the method for identifying and warning of stress-induced hyperglycemia as described in any one of claims 1 to 6.

Images