A method for risk data evaluation analysis of sepsis acute kidney injury

Abstract

The present application relates to the field of risk data evaluation analysis of sepsis acute kidney injury, and particularly relates to a risk data evaluation analysis method for sepsis acute kidney injury, comprising: establishing associated data multi-level features of sepsis acute kidney injury by using associated data of sepsis acute kidney injury; establishing a data risk analysis model of sepsis acute kidney injury based on deep learning by using the associated data multi-level features of sepsis acute kidney injury; and obtaining a risk data evaluation analysis result through output consistency verification processing according to the data risk analysis model of sepsis acute kidney injury, wherein the risk evaluation process with self-adaptive optimization capability is constructed through dynamic feature fusion of multi-source medical data and deep learning modeling, the prediction accuracy and timeliness are significantly improved compared with traditional analysis methods, and the result reliability is ensured through the output verification mechanism.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of risk data evaluation analysis of sepsis acute kidney injury, and in particular to a risk data evaluation analysis method for sepsis acute kidney injury. BACKGROUND

[0002] Sepsis is a life-threatening organ dysfunction caused by a dysregulated host response to infection, and its complication, acute kidney injury, significantly increases patient mortality and medical burden. The pathophysiological mechanism of sepsis-related acute kidney injury is complex, involving inflammation, hemodynamic disorders, microcirculatory disorders, and other factors interacting with each other, which poses great challenges to early diagnosis and risk assessment. Currently, the risk assessment of SA-AKI in clinical practice mainly relies on single-time-point static indicators (such as serum creatinine, urine volume, etc.) or traditional scoring systems. However, these methods have the following limitations: Insufficient sensitivity of static indicators: indicators such as serum creatinine usually increase only after significant impairment of renal function, resulting in delayed early warning. Ignoring the dynamic trend characteristics: the existing methods do not fully integrate the time series variation of patient signs (such as heart rate, arterial pressure) and laboratory indicators (such as lactic acid, anion gap), making it difficult to reflect the dynamic progression of the disease. Lack of multidimensional correlation analysis: traditional models rarely explore the interaction between patient signs and laboratory indicators (such as the correlation between heart rate fluctuations and elevated blood lactic acid), resulting in limited risk prediction accuracy. Insufficient generalization ability: clinical data have individual differences and noise interference, and existing models are prone to failure due to data distribution shift, requiring dynamic updating to improve robustness. In recent years, deep learning technology has shown advantages in medical data analysis, which can improve prediction performance through multi-level feature extraction and complex pattern recognition. However, existing researches mostly focus on single data type (such as only using laboratory indicators or image data), lack of collaborative modeling of multi-source heterogeneous data, and model validation often ignores clinical consistency. SUMMARY

[0003] In view of the deficiencies of the prior art, the present application provides a risk data evaluation analysis method for sepsis acute kidney injury, which integrates multi-level features of patient sign data and laboratory indicator data, constructs and optimizes a risk analysis model based on deep learning, and then performs output consistency verification processing to achieve dynamic and accurate risk assessment of sepsis-related acute kidney injury.

[0004] To achieve the above-mentioned purpose, the present application provides a risk data evaluation analysis method for sepsis acute kidney injury, comprising:

[0005] S1, establishing multi-level features of sepsis acute kidney injury associated data using sepsis acute kidney injury associated data;

[0006] S2, establish a data risk analysis model of sepsis acute kidney injury based on deep learning using the multi-level features of the correlation data of sepsis acute kidney injury;

[0007] S3, perform output consistency verification processing according to the data risk analysis model of sepsis acute kidney injury to obtain a risk data evaluation analysis result.

[0008] Preferably, the multi-level features of the correlation data of sepsis acute kidney injury include:

[0009] S1-1, collect patient sign data and laboratory index data of sepsis acute kidney injury as correlation data of sepsis acute kidney injury;

[0010] S1-2, establish a correlation data numerical feature of sepsis acute kidney injury using the correlation data of sepsis acute kidney injury;

[0011] S1-3, establish a correlation data trend feature of sepsis acute kidney injury using the correlation data of sepsis acute kidney injury;

[0012] S1-4, use the correlation data numerical feature and the correlation data trend feature as the multi-level features of the correlation data of sepsis acute kidney injury;

[0013] The patient sign data includes heart rate and arterial pressure, and the laboratory index data includes anion gap, carbon dioxide combining power, blood urea nitrogen, blood creatinine, blood potassium, blood lactic acid, and hemoglobin.

[0014] Further, establishing a correlation data numerical feature of sepsis acute kidney injury using the correlation data of sepsis acute kidney injury includes:

[0015] S1-2-1, obtain historical patient sign data according to the patient sign data corresponding to the correlation data of sepsis acute kidney injury;

[0016] S1-2-2, obtain historical laboratory index data according to the laboratory index data corresponding to the correlation data of sepsis acute kidney injury;

[0017] S1-2-3, obtain historical maximum heart rate and historical minimum heart rate corresponding to the historical patient sign data, respectively;

[0018] S1-2-4, obtain historical maximum arterial pressure and historical minimum arterial pressure corresponding to the historical patient sign data, respectively;

[0019] S1-2-5, obtain a difference between the historical maximum heart rate and the historical minimum heart rate as a historical heart rate fluctuation value;

[0020] S1-2-6, obtaining the difference between the historical arterial pressure maximum value and the historical arterial pressure minimum value as a historical arterial pressure floating value;

[0021] S1-2-7, establishing a heart rate-arterial pressure interval mapping using the historical heart rate floating value and the historical arterial pressure floating value;

[0022] S1-2-8, obtaining historical heart rate maximum value corresponding assay index data as heart rate peak value auxiliary data according to the historical heart rate maximum value respectively;

[0023] S1-2-9, obtaining historical heart rate minimum value corresponding assay index data as heart rate valley value auxiliary data according to the historical heart rate minimum value respectively;

[0024] S1-2-10, obtaining historical arterial pressure maximum value corresponding assay index data as arterial pressure peak value auxiliary data according to the historical arterial pressure maximum value respectively;

[0025] S1-2-11, obtaining historical arterial pressure minimum value corresponding assay index data as arterial pressure valley value auxiliary data according to the historical arterial pressure minimum value respectively;

[0026] S1-2-12, using the heart rate-arterial pressure interval mapping, heart rate peak value auxiliary data, heart rate valley value auxiliary data, arterial pressure peak value auxiliary data and arterial pressure valley value auxiliary data as the correlation data numerical characteristics of sepsis acute kidney injury.

[0027] Further, using the correlation data of sepsis acute kidney injury to establish the correlation data trend characteristics of sepsis acute kidney injury includes:

[0028] Using the correlation data of sepsis acute kidney injury corresponding to the time as the standard starting time t;

[0029] Obtaining the patient sign data corresponding to the correlation data of sepsis acute kidney injury at t-1 time;

[0030] Obtaining the patient sign data corresponding to the correlation data of sepsis acute kidney injury at t+1 time;

[0031] Obtaining the assay index data corresponding to the correlation data of sepsis acute kidney injury at t-1 time;

[0032] Obtaining the assay index data corresponding to the correlation data of sepsis acute kidney injury at t+1 time;

[0033] Respectively obtaining the data change trend of the patient sign data corresponding to the patient sign data at t-1 time to the standard starting time t, the standard starting time t to t+1 time as the patient sign data trend;

[0034] The data change trends of the test indicators from time t-1 to the standard start time t and from the standard start time t to time t+1 are respectively obtained as the test indicator data trends;

[0035] The trends in patient vital signs and laboratory indicators were used as correlation trend features for acute kidney injury in sepsis.

[0036] The data change trend refers to the changes in adjacent data.

[0037] Furthermore, a data risk analysis model for septic acute kidney injury (SKI) is established based on deep learning using the multi-level features of the associated data, including:

[0038] S2-1. Using the multi-level features of the association data of acute kidney injury in sepsis, a feature association analysis model of acute kidney injury in sepsis is established based on deep learning;

[0039] S2-2. Using the feature association analysis model of sepsis-induced acute kidney injury, a data risk analysis model of sepsis-induced acute kidney injury is obtained through generalization verification and improvement processing.

[0040] Furthermore, utilizing the multi-level features of the association data of septic acute kidney injury, a feature association analysis model for septic acute kidney injury is established based on deep learning, including:

[0041] S2-1-1. Establish a first dataset using the patient's vital signs data;

[0042] S2-1-2. Establish a second dataset using the aforementioned laboratory indicator data;

[0043] S2-1-3. A third dataset is established using the trend characteristics of patient vital signs data corresponding to the multi-level features of the associated data of the sepsis acute kidney injury.

[0044] S2-1-4. A fourth dataset is established by using the multi-level features of the correlation data of the sepsis acute kidney injury to correspond to the trend features of the laboratory index data trends of the correlation data trends.

[0045] S2-1-5. Using the first dataset as input and the third dataset corresponding to the first dataset as output, a data association model of the first patient's vital signs is established by training based on deep learning.

[0046] S2-1-6. Using the first dataset as input and the fourth dataset corresponding to the first dataset as output, a second patient vital sign data association model is established by training and processing based on deep learning.

[0047] S2-1-7, training and establishing a first laboratory index data correlation model based on deep learning by taking the second data set as input and the third data set corresponding to the second data set as output;

[0048] S2-1-8, training and establishing a second laboratory index data correlation model based on deep learning by taking the second data set as input and the fourth data set corresponding to the second data set as output;

[0049] S2-1-9, using the first patient sign data correlation model, the second patient sign data correlation model, the first laboratory index data correlation model and the second laboratory index data correlation model as a feature correlation analysis model of sepsis acute kidney injury.

[0050] Further, the sepsis acute kidney injury data risk analysis model is improved by generalization verification using the feature correlation analysis model of sepsis acute kidney injury.

[0051] S2-2-1, obtaining all first patient sign data correlation results by taking the first data set as input and the first patient sign data correlation model corresponding to the feature correlation analysis model of sepsis acute kidney injury as output;

[0052] S2-2-2, obtaining all second patient sign data correlation results by taking the first data set as input and the second patient sign data correlation model corresponding to the feature correlation analysis model of sepsis acute kidney injury as output;

[0053] S2-2-3, obtaining all first laboratory index data correlation results by taking the second data set as input and the first laboratory index data correlation model corresponding to the feature correlation analysis model of sepsis acute kidney injury as output;

[0054] S2-2-4, obtaining all second laboratory index data correlation results by taking the second data set as input and the second laboratory index data correlation model corresponding to the feature correlation analysis model of sepsis acute kidney injury as output;

[0055] S2-2-5, determining whether the first data set is located in the heart rate-arterial pressure interval mapping of the correlation data value characteristics of sepsis acute kidney injury, if yes, executing S2-2-6, otherwise, updating the historical patient sign data by taking the first data set corresponding to the subset located outside the heart rate-arterial pressure interval mapping, and returning to execute S1-2-2;

[0056] S2-2-6, determining whether the overall first patient sign data association result exceeds the range of the heart rate peak value auxiliary data and the heart rate valley value auxiliary data of the association data numerical characteristic of sepsis acute kidney injury, if so, updating the historical patient sign data with the patient sign data corresponding to the first patient sign data association result exceeding the range of the heart rate peak value auxiliary data and the heart rate valley value auxiliary data, and returning to execute S1-2-2, otherwise, executing S2-2-7;

[0057] S2-2-7, determining whether the overall second patient sign data association result exceeds the range of the arterial pressure peak value auxiliary data and the arterial pressure valley value auxiliary data of the association data numerical characteristic of sepsis acute kidney injury, if so, updating the historical laboratory index data with the laboratory index data corresponding to the second patient sign data association result exceeding the range of the arterial pressure peak value auxiliary data and the arterial pressure valley value auxiliary data, and returning to execute S1-2-3, otherwise, executing S2-2-8;

[0058] S2-2-8, determining whether the overall first laboratory index data association result exceeds the range of the heart rate peak value auxiliary data and the heart rate valley value auxiliary data of the association data numerical characteristic of sepsis acute kidney injury, if so, updating the historical patient sign data with the patient sign data corresponding to the first laboratory index data association result exceeding the range of the heart rate peak value auxiliary data and the heart rate valley value auxiliary data, and returning to execute S1-2-2, otherwise, executing S2-2-9;

[0059] S2-2-9, determining whether the overall second laboratory index data association result exceeds the range of the heart rate peak value auxiliary data and the heart rate valley value auxiliary data of the association data numerical characteristic of sepsis acute kidney injury, if so, updating the historical laboratory index data with the laboratory index data corresponding to the second laboratory index data association result exceeding the range of the heart rate peak value auxiliary data and the heart rate valley value auxiliary data, and returning to execute S1-2-3, otherwise, using the sepsis acute kidney injury feature association analysis model as a data risk analysis model of sepsis acute kidney injury.

[0060] Further, the output consistency verification processing according to the data risk analysis model of sepsis acute kidney injury obtains a risk data evaluation analysis result, which includes:

[0061] S3-1, obtaining a basic data risk analysis result of sepsis acute kidney injury by using the data risk analysis model of sepsis acute kidney injury;

[0062] S3-2, performing output consistency verification processing to obtain a risk data evaluation analysis result by using the basic data risk analysis result of sepsis acute kidney injury.

[0063] Further, the basic data risk analysis result of the sepsis acute kidney injury is obtained by using the data risk analysis model of the sepsis acute kidney injury, and the basic data risk analysis result of the sepsis acute kidney injury includes:

[0064] The real-time first patient sign data correlation result, the real-time second patient sign data correlation result, the real-time first laboratory index data correlation result and the real-time second laboratory index data correlation result are obtained by inputting the patient sign data and the laboratory index data into the data risk analysis model of the sepsis acute kidney injury respectively;

[0065] The laboratory index data trend corresponding to the historical first patient sign data correlation result and the historical second patient sign data correlation result is obtained as the to-be-analyzed laboratory index data trend according to the real-time first patient sign data correlation result and the real-time second patient sign data correlation result;

[0066] The patient sign data trend corresponding to the historical first laboratory index data correlation result and the historical second laboratory index data correlation result is obtained as the to-be-analyzed patient sign data trend according to the real-time first laboratory index data correlation result and the real-time second laboratory index data correlation result;

[0067] The to-be-analyzed laboratory index data trend and the to-be-analyzed patient sign data trend are used as the basic data risk analysis result of the sepsis acute kidney injury.

[0068] Further, the risk data evaluation analysis result is obtained by performing output consistency verification processing on the basic data risk analysis result of the sepsis acute kidney injury, and the risk data evaluation analysis result includes:

[0069] S3-2-1, it is judged whether the basic data risk analysis result of the sepsis acute kidney injury corresponding to the to-be-analyzed laboratory index data trend and the real-time first laboratory index data correlation result is consistent in trend, if yes, S3-2-2 is executed, otherwise, the laboratory index data corresponding to the real-time first laboratory index data correlation result is output as the risk data evaluation analysis result;

[0070] S3-2-2, it is judged whether the basic data risk analysis result of the sepsis acute kidney injury corresponding to the to-be-analyzed laboratory index data trend and the real-time second laboratory index data correlation result is consistent in trend, if yes, S3-2-3 is executed, otherwise, the laboratory index data corresponding to the real-time second laboratory index data correlation result is output as the risk data evaluation analysis result;

[0071] S3-2-3, it is judged whether the basic data risk analysis result of the sepsis acute kidney injury corresponding to the to-be-analyzed patient sign data trend and the real-time first patient sign data correlation result is consistent in trend, if yes, S3-2-4 is executed, otherwise, the patient sign data corresponding to the real-time first patient sign data correlation result is output as the risk data evaluation analysis result;

[0072] S3-2-4, judge whether the trend consistency of the basis data risk analysis result of the sepsis acute kidney injury corresponds to the trend of the patient sign data and the real-time first patient sign data association result, if yes, update the association data of the sepsis acute kidney injury, and return to S1-2, otherwise, output the patient sign data corresponding to the real-time second patient sign data association result as the risk data evaluation analysis result;

[0073] The trend consistency is the same change of adjacent data before and after.

[0074] Compared with the closest prior art, the present application has the beneficial effects of:

[0075] Early warning capability is improved: by fusing multi-level features of patient sign data and laboratory index data, early risk signals of sepsis acute kidney injury can be more sensitively captured, making up for the lagging defects of traditional static indicators;

[0076] Enhance risk assessment accuracy: the data risk analysis model based on deep learning can mine the complex association between sign data and laboratory indicators (such as the cooperative change rule of heart rate fluctuation and blood lactic acid rise), improve prediction accuracy, and reduce misjudgment and omission;

[0077] Dynamic adaptability optimization: through the generalization verification improvement mechanism, the model can adapt to individual differences and dynamic changes of clinical data, improve robustness and generalization ability;

[0078] Strong clinical interpretability: trend consistency verification is adopted to ensure that the model output conforms to the medical logic, enhance the trust of doctors on the results, and facilitate clinical decision-making;

[0079] Reduce medical risk: through real-time monitoring and dynamic evaluation, high-risk patients can be warned in advance, valuable time is saved for clinical intervention, and the risk of kidney function deterioration and multiple organ failure is reduced.

[0080] Automation and intelligence: reduce manual dependence, improve evaluation efficiency, suitable for fast-paced medical scenes such as ICU and emergency department, and help precise medicine and individualized treatment. BRIEF DESCRIPTION OF DRAWINGS

[0081] Figure 1 is a flowchart of a risk data evaluation analysis method for sepsis acute kidney injury provided by the present application. DETAILED DESCRIPTION

[0082] The specific embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

[0083] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0084] Embodiment 1

[0085] The present application provides a risk data evaluation analysis method for sepsis acute kidney injury, as shown in the formula (I), comprising: Figure 1

[0086] S1, establishing a sepsis acute kidney injury correlation data multi-level feature by using sepsis acute kidney injury correlation data;

[0087] S2, establishing a sepsis acute kidney injury data risk analysis model based on deep learning by using the sepsis acute kidney injury correlation data multi-level feature;

[0088] S3, performing output consistency verification processing to obtain a risk data evaluation analysis result according to the sepsis acute kidney injury data risk analysis model.

[0089] S1 specifically comprises:

[0090] S1-1, collecting patient sign data and laboratory index data of sepsis acute kidney injury as sepsis acute kidney injury correlation data respectively;

[0091] S1-2, establishing a sepsis acute kidney injury correlation data numerical feature by using the sepsis acute kidney injury correlation data;

[0092] S1-3, establishing a sepsis acute kidney injury correlation data trend feature by using the sepsis acute kidney injury correlation data;

[0093] S1-4, using the sepsis acute kidney injury correlation data numerical feature and the sepsis acute kidney injury correlation data trend feature as a sepsis acute kidney injury correlation data multi-level feature;

[0094] The patient sign data comprises heart rate and arterial pressure, and the laboratory index data comprises anion gap, carbon dioxide combining power, blood urea nitrogen, blood creatinine, blood potassium, blood lactic acid and hemoglobin.

[0095] S1-2 specifically comprises:

[0096] S1-2-1, obtaining historical patient sign data according to the sepsis acute kidney injury correlation data corresponding to the patient sign data; ​

[0097] S1-2-2, acquiring historical assay index data corresponding to the sepsis acute kidney injury correlation data;

[0098] S1-2-3, acquiring historical maximum heart rate values and historical minimum heart rate values corresponding to the historical patient sign data respectively;

[0099] S1-2-4, acquiring historical maximum arterial pressure values and historical minimum arterial pressure values corresponding to the historical patient sign data respectively;

[0100] S1-2-5, acquiring the difference between the historical maximum heart rate value and the historical minimum heart rate value as a historical heart rate fluctuation value;

[0101] S1-2-6, acquiring the difference between the historical maximum arterial pressure value and the historical minimum arterial pressure value as a historical arterial pressure fluctuation value;

[0102] S1-2-7, establishing a heart rate-arterial pressure interval mapping using the historical heart rate fluctuation value and the historical arterial pressure fluctuation value;

[0103] S1-2-8, acquiring historical assay index data corresponding to the historical maximum heart rate value as heart rate peak value auxiliary data according to the historical maximum heart rate value respectively;

[0104] S1-2-9, acquiring historical assay index data corresponding to the historical minimum heart rate value as heart rate valley value auxiliary data according to the historical minimum heart rate value respectively;

[0105] S1-2-10, acquiring historical assay index data corresponding to the historical maximum arterial pressure value as arterial pressure peak value auxiliary data according to the historical maximum arterial pressure value respectively;

[0106] S1-2-11, acquiring historical assay index data corresponding to the historical minimum arterial pressure value as arterial pressure valley value auxiliary data according to the historical minimum arterial pressure value respectively;

[0107] S1-2-12, using the heart rate-arterial pressure interval mapping, the heart rate peak value auxiliary data, the heart rate valley value auxiliary data, the arterial pressure peak value auxiliary data, and the arterial pressure valley value auxiliary data as the sepsis acute kidney injury correlation data numerical features.

[0108] S1-3 specifically includes:

[0109] S1-3-1, using the time corresponding to the sepsis acute kidney injury correlation data as a standard starting time t;

[0110] S1-3-2, acquiring patient sign data corresponding to the sepsis acute kidney injury correlation data at time t-1;

[0111] S1-3-3, obtain the patient sign data corresponding to the correlation data of sepsis acute kidney injury at t+1 time;

[0112] S1-3-4, obtain the test index data corresponding to the correlation data of sepsis acute kidney injury at t-1 time;

[0113] S1-3-5, obtain the test index data corresponding to the correlation data of sepsis acute kidney injury at t+1 time;

[0114] S1-3-6, respectively obtain the data change trend of the patient sign data from t-1 time to standard starting time t, and from standard starting time t to t+1 time as the patient sign data trend;

[0115] S1-3-7, respectively obtain the data change trend of the test index data from t-1 time to standard starting time t, and from standard starting time t to t+1 time as the test index data trend;

[0116] S1-3-8, use the patient sign data trend and the test index data trend as the correlation data trend characteristics of sepsis acute kidney injury;

[0117] Wherein, the data change trend is the change of adjacent data.

[0118] S2 specifically includes:

[0119] S2-1, use the correlation data multi-level characteristics of sepsis acute kidney injury to establish a feature correlation analysis model of sepsis acute kidney injury based on deep learning;

[0120] S2-2, use the feature correlation analysis model of sepsis acute kidney injury to carry out generalization verification and improvement processing to obtain a data risk analysis model of sepsis acute kidney injury.

[0121] S2-1 specifically includes:

[0122] S2-1-1, use the patient sign data to establish a first data set;

[0123] S2-1-2, use the test index data to establish a second data set;

[0124] S2-1-3, use the patient sign data trend corresponding to the correlation data trend characteristics of the correlation data multi-level characteristics of sepsis acute kidney injury to establish a third data set;

[0125] S2-1-4, use the test index data trend corresponding to the correlation data trend characteristics of the correlation data multi-level characteristics of sepsis acute kidney injury to establish a fourth data set;

[0126] S2-1-5, training and establishing a first patient sign data correlation model based on deep learning by taking the first data set as input and the third data set corresponding to the first data set as output;

[0127] S2-1-6, training and establishing a second patient sign data correlation model based on deep learning by taking the first data set as input and the fourth data set corresponding to the first data set as output;

[0128] S2-1-7, training and establishing a first laboratory index data correlation model based on deep learning by taking the second data set as input and the third data set corresponding to the second data set as output;

[0129] S2-1-8, training and establishing a second laboratory index data correlation model based on deep learning by taking the second data set as input and the fourth data set corresponding to the second data set as output;

[0130] S2-1-9, taking the first patient sign data correlation model, the second patient sign data correlation model, the first laboratory index data correlation model and the second laboratory index data correlation model as a feature correlation analysis model of sepsis acute kidney injury.

[0131] S2-2 specifically includes:

[0132] S2-2-1, obtaining all first patient sign data correlation results by taking the first data set and the first patient sign data correlation model corresponding to the feature correlation analysis model of sepsis acute kidney injury;

[0133] S2-2-2, obtaining all second patient sign data correlation results by taking the first data set and the second patient sign data correlation model corresponding to the feature correlation analysis model of sepsis acute kidney injury;

[0134] S2-2-3, obtaining all first laboratory index data correlation results by taking the second data set and the first laboratory index data correlation model corresponding to the feature correlation analysis model of sepsis acute kidney injury;

[0135] S2-2-4, obtaining all second laboratory index data correlation results by taking the second data set and the second laboratory index data correlation model corresponding to the feature correlation analysis model of sepsis acute kidney injury;

[0136] S2-2-5, judging whether the first data set is located in the heart rate-arterial pressure interval mapping of the correlation data value characteristics of sepsis acute kidney injury, if yes, executing S2-2-6, otherwise, updating the historical patient sign data by taking the first data set corresponding to the subset located outside the heart rate-arterial pressure interval mapping, and returning to execute S1-2-2;

[0137] S2-2-6, determining whether the overall first patient sign data association result exceeds the range of the heart rate peak value auxiliary data and the heart rate valley value auxiliary data of the association data numerical characteristic of sepsis acute kidney injury, if so, updating the historical patient sign data with the patient sign data corresponding to the first patient sign data association result exceeding the range of the heart rate peak value auxiliary data and the heart rate valley value auxiliary data, and returning to execute S1-2-2, otherwise, executing S2-2-7;

[0138] S2-2-7, determining whether the overall second patient sign data association result exceeds the range of the arterial pressure peak value auxiliary data and the arterial pressure valley value auxiliary data of the association data numerical characteristic of sepsis acute kidney injury, if so, updating the historical laboratory index data with the laboratory index data corresponding to the second patient sign data association result exceeding the range of the arterial pressure peak value auxiliary data and the arterial pressure valley value auxiliary data, and returning to execute S1-2-3, otherwise, executing S2-2-8;

[0139] S2-2-8, determining whether the overall first laboratory index data association result exceeds the range of the heart rate peak value auxiliary data and the heart rate valley value auxiliary data of the association data numerical characteristic of sepsis acute kidney injury, if so, updating the historical patient sign data with the patient sign data corresponding to the first laboratory index data association result exceeding the range of the heart rate peak value auxiliary data and the heart rate valley value auxiliary data, and returning to execute S1-2-2, otherwise, executing S2-2-9;

[0140] S2-2-9, determining whether the overall second laboratory index data association result exceeds the range of the heart rate peak value auxiliary data and the heart rate valley value auxiliary data of the association data numerical characteristic of sepsis acute kidney injury, if so, updating the historical laboratory index data with the laboratory index data corresponding to the second laboratory index data association result exceeding the range of the heart rate peak value auxiliary data and the heart rate valley value auxiliary data, and returning to execute S1-2-3, otherwise, using the sepsis acute kidney injury feature association analysis model as a data risk analysis model for sepsis acute kidney injury.

[0141] S3 specifically comprises:

[0142] S3-1, obtaining a basic data risk analysis result of sepsis acute kidney injury by using the data risk analysis model for sepsis acute kidney injury;

[0143] S3-2, performing output consistency verification processing by using the basic data risk analysis result of sepsis acute kidney injury to obtain a risk data evaluation analysis result.

[0144] S3-1 specifically comprises:

[0145] S3-1-1, inputting the patient sign data and the laboratory index data into a data risk analysis model of sepsis acute kidney injury to obtain real-time first patient sign data correlation results, real-time second patient sign data correlation results, real-time first laboratory index data correlation results and real-time second laboratory index data correlation results respectively;

[0146] S3-1-2, acquiring a laboratory index data trend corresponding to the historical first patient sign data correlation results and the historical second patient sign data correlation results according to the real-time first patient sign data correlation results and the real-time second patient sign data correlation results as a to-be-analyzed laboratory index data trend;

[0147] S3-1-3, acquiring a patient sign data trend corresponding to the historical first laboratory index data correlation results and the historical second laboratory index data correlation results according to the real-time first laboratory index data correlation results and the real-time second laboratory index data correlation results as a to-be-analyzed patient sign data trend;

[0148] S3-1-4, using the to-be-analyzed laboratory index data trend and the to-be-analyzed patient sign data trend as a basic data risk analysis result of sepsis acute kidney injury.

[0149] S3-2 specifically comprises:

[0150] S3-2-1, judging whether the basic data risk analysis result of sepsis acute kidney injury corresponding to the to-be-analyzed laboratory index data trend and the real-time first laboratory index data correlation results are consistent in trend, if yes, executing S3-2-2, otherwise, outputting the laboratory index data corresponding to the real-time first laboratory index data correlation results as a risk data evaluation analysis result;

[0151] S3-2-2, judging whether the basic data risk analysis result of sepsis acute kidney injury corresponding to the to-be-analyzed laboratory index data trend and the real-time second laboratory index data correlation results are consistent in trend, if yes, executing S3-2-3, otherwise, outputting the laboratory index data corresponding to the real-time second laboratory index data correlation results as a risk data evaluation analysis result;

[0152] S3-2-3, judging whether the basic data risk analysis result of sepsis acute kidney injury corresponding to the to-be-analyzed patient sign data trend and the real-time first patient sign data correlation results are consistent in trend, if yes, executing S3-2-4, otherwise, outputting the patient sign data corresponding to the real-time first patient sign data correlation results as a risk data evaluation analysis result;

[0153] S3-2-4, judging whether the trend of the basis data risk analysis result of the sepsis acute kidney injury corresponds to the trend of the real-time first patient sign data association result, if yes, updating the association data of the sepsis acute kidney injury, and returning to S1-2, otherwise, outputting the real-time second patient sign data association result as the risk data evaluation analysis result;

[0154] The trend consistency is that the change of adjacent data is the same.

[0155] In this embodiment, a risk data evaluation analysis method for sepsis acute kidney injury is provided, and the specific implementation is as follows:

[0156] 1. Data collection and preprocessing:

[0157] Data source: sepsis patients (meeting the Sepsis-3.0 diagnostic criteria) admitted to the ICU of a certain third-grade class-A hospital from 2020 to 2023 were selected, a total of 1,200 cases were included, of which 480 cases (40%) were patients with AKI (KDIGO standard).

[0158] Data content:

[0159] Patient sign data: heart rate (HR), mean arterial pressure (MAP), recorded every 1 hour.

[0160] Laboratory index data: blood lactic acid (Lac), serum creatinine (Scr), blood urea nitrogen (BUN), potassium (K⁺), hemoglobin (Hb), detected every 12 or 24 hours.

[0161] Data cleaning: exclude cases with a data missing rate of >20%, and perform medical reasonableness check on abnormal values (such as Scr>1,000 μmol / L).

[0162] 2. Multi-level feature extraction:

[0163] Numerical feature construction (S1-2):

[0164] Calculate the historical heart rate fluctuation value (HR_max - HR_min) and the historical arterial pressure fluctuation value (MAP_max - MAP_min).

[0165] Establish a heart rate-arterial pressure interval mapping (such as HR 60-100 times / min + MAP 65-110 mmHg as a safe interval).

[0166] Extract peak / valley auxiliary data (such as the highest HR corresponding Lac, Scr value).

[0167] Trend feature construction (S1-3):

[0168] Compute 12 or 24 hours trend (e.g. Scr rising rate > 26.5 pmol / L / 6h is considered high risk).

[0169] Compute blood lactate slope (Lac change per hour > 0.5 mmol / L is considered worsening trend).

[0170] 3. Deep learning model training:

[0171] Model architecture: Adopt multi-modal neural network (MM-Net), including:

[0172] Input layer: Vital signs data (HR, MAP) + Lab data (Lac, Scr, BUN).

[0173] Feature fusion layer: Use attention mechanism (Attention) to weight the importance of different features.

[0174] Output layer: SA-AKI risk probability (0-1).

[0175] Training strategy:

[0176] Dataset division: Training set (70%), validation set (15%), test set (15%).

[0177] Optimization method: Adam optimizer, learning rate 0.001, early stopping method (patience=10).

[0178] Performance evaluation:

[0179] AUC = 0.91 (95% CI: 0.88-0.93), better than traditional Logistic regression model (AUC=0.82).

[0180] Sensitivity = 85.2%, specificity = 79.6%.

[0181] 4. Dynamic risk early warning:

[0182] Real-time monitoring case:

[0183] Patient A (septic shock, initial Scr=80 pmol / L):

[0184] Input data: HR=125 times / min (out of safe interval), Lac 2.1→4.8 mmol / L (within 3h).

[0185] Model output: SA-AKI risk probability = 92% (high risk).

[0186] Clinical intervention: Adjust vasoactive drugs + optimize fluid management, eventually no AKI (Scr stable).

[0187] Patient B (initially normal Scr, but abnormal trend):

[0188] Input data: BUN increased by >5 mmol / L within 12h or 24h, MAP fluctuation >30%.

[0189] Model output: SA-AKI risk probability = 87% (medium-high risk).

[0190] Clinical outcome: Confirmed AKI (KDIGO stage 2) after 48h.

[0191] Among them, the generalization ability verification process of the overall model includes:

[0192] 1. External validation dataset:

[0193] Another hospital's 2023-2024 data (n=500, SA-AKI incidence rate 38%).

[0194] Model performance:

[0195] AUC = 0.88 (95% CI: 0.85-0.90), sensitivity = 82.4%.

[0196] In patients with low MAP (<65 mmHg), the prediction accuracy still remains above 85%.

[0197] 2. Adaptive updating mechanism:

[0198] Abnormal data identification: When the proportion of HR > 140 times / min in new data is 20% higher than that in the training set, the model update is triggered.

[0199] Incremental learning: After adding 300 new data, the model AUC is improved to 0.89 (originally 0.87).

[0200] Those skilled in the art will appreciate that embodiments of the present application can be provided as methods, systems, or computer program products. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage media, etc.) having computer-usable program code embodied therein.

[0201] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart or flows and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart or flows and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart or flows and / or block diagram block or blocks.

[0202] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flowchart or flows and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart or flows and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart or flows and / or block diagram block or blocks.

[0203] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart or flows and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart or flows and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart or flows and / or block diagram block or blocks.

[0204] Finally, it should be noted that the above-mentioned embodiments are merely intended to illustrate the technical solutions of the present application, but not to limit it. Although the present application has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that the technical solutions of the present application can still be modified or equivalent replaced without departing from the spirit and scope of the present application, and any modification or equivalent replacement should be covered in the protection scope of the claims of the present application.

Claims

1. A method for risk data evaluation analysis of sepsis acute kidney injury, characterized in that, Comprising: S1, using the correlation data of sepsis acute kidney injury to establish the correlation data multi-level characteristics of sepsis acute kidney injury; S1-1, respectively collecting the patient sign data and the test index data of sepsis acute kidney injury as the correlation data of sepsis acute kidney injury; S1-2, using the correlation data of sepsis acute kidney injury to establish the correlation data numerical characteristics of sepsis acute kidney injury; S1-3, using the correlation data of sepsis acute kidney injury to establish the correlation data trend characteristics of sepsis acute kidney injury; S1-4, using the correlation data numerical characteristics and the correlation data trend characteristics of sepsis acute kidney injury as the correlation data multi-level characteristics of sepsis acute kidney injury; Wherein, the patient sign data includes heart rate and arterial pressure, and the test index data includes anion gap, carbon dioxide combining power, blood urea nitrogen, blood creatinine, blood potassium, blood lactic acid and hemoglobin; S2, using the correlation data multi-level characteristics of sepsis acute kidney injury to establish the data risk analysis model of sepsis acute kidney injury based on deep learning; S2-1, using the correlation data multi-level characteristics of sepsis acute kidney injury to establish the feature correlation analysis model of sepsis acute kidney injury based on deep learning; S2-1-1, using the patient sign data to establish a first data set; S2-1-2, using the test index data to establish a second data set; S2-1-3, using the patient sign data trend corresponding to the correlation data trend characteristics of the correlation data multi-level characteristics of sepsis acute kidney injury to establish a third data set; S2-1-4, using the test index data trend corresponding to the correlation data trend characteristics of the correlation data multi-level characteristics of sepsis acute kidney injury to establish a fourth data set; S2-1-5, using the first data set as input and the third data set corresponding to the first data set as output, training based on deep learning to establish a first patient sign data correlation model; S2-1-6, using the first data set as input and the fourth data set corresponding to the first data set as output, training based on deep learning to establish a second patient sign data correlation model; S2-1-7, using the second data set as input and the third data set corresponding to the second data set as output, training based on deep learning to establish a first test index data correlation model; S2-1-8, using the second data set as input and the fourth data set corresponding to the second data set as output, training based on deep learning to establish a second test index data correlation model; S2-1-9, using the first patient sign data correlation model, the second patient sign data correlation model, the first test index data correlation model and the second test index data correlation model as the feature correlation analysis model of sepsis acute kidney injury; S2-2, using the feature correlation analysis model of sepsis acute kidney injury to carry out generalization verification and improvement processing to obtain the data risk analysis model of sepsis acute kidney injury; S2-2-1, obtaining all first patient sign data association results according to the first data set and a first patient sign data association model corresponding to a feature association analysis model of sepsis acute kidney injury; S2-2-2, obtaining all second patient sign data association results according to the first data set and a second patient sign data association model corresponding to the feature association analysis model of sepsis acute kidney injury; S2-2-3, obtaining all first laboratory index data association results according to the second data set and a first laboratory index data association model corresponding to the feature association analysis model of sepsis acute kidney injury; S2-2-4, obtaining all second laboratory index data association results according to the second data set and a second laboratory index data association model corresponding to the feature association analysis model of sepsis acute kidney injury; S2-2-5, determining whether the first data set is located in a heart rate-arterial pressure interval mapping of the associated data numerical characteristics of sepsis acute kidney injury, if yes, executing S2-2-6, otherwise, updating the historical patient sign data by using a subset corresponding to the first data set located outside the heart rate-arterial pressure interval mapping, and returning to execute S1-2-2; S2-2-6, determining whether the all first patient sign data association results exceed the range of heart rate peak auxiliary data and heart rate valley auxiliary data of the associated data numerical characteristics of sepsis acute kidney injury, if yes, updating the historical patient sign data by using the patient sign data corresponding to the first patient sign data association results exceeding the range of heart rate peak auxiliary data and heart rate valley auxiliary data, and returning to execute S1-2-2, otherwise, executing S2-2-7; S2-2-7, determining whether the all second patient sign data association results exceed the range of arterial pressure peak auxiliary data and arterial pressure valley auxiliary data of the associated data numerical characteristics of sepsis acute kidney injury, if yes, updating the historical laboratory index data by using the laboratory index data corresponding to the second patient sign data association results exceeding the range of arterial pressure peak auxiliary data and arterial pressure valley auxiliary data, and returning to execute S1-2-3, otherwise, executing S2-2-8; S2-2-8, determining whether the all first laboratory index data association results exceed the range of heart rate peak auxiliary data and heart rate valley auxiliary data of the associated data numerical characteristics of sepsis acute kidney injury, if yes, updating the historical patient sign data by using the patient sign data corresponding to the first laboratory index data association results exceeding the range of heart rate peak auxiliary data and heart rate valley auxiliary data, and returning to execute S1-2-2, otherwise, executing S2-2-9; S2-2-9, judging whether the overall second test index data correlation result exceeds the range of the heart rate peak value auxiliary data and the heart rate valley value auxiliary data of the correlation data numerical characteristics of sepsis acute kidney injury, if so, updating the historical test index data with the test index data correlation result exceeding the range of the heart rate peak value auxiliary data and the heart rate valley value auxiliary data, and returning to execute S1-2-3, otherwise, using the feature correlation analysis model of sepsis acute kidney injury as a data risk analysis model of sepsis acute kidney injury; S3, according to the data risk analysis model of sepsis acute kidney injury, performing output consistency verification processing to obtain a risk data evaluation analysis result.

2. A method for risk data evaluation analysis of sepsis acute kidney injury as claimed in claim 1 wherein, The correlation data numerical characteristics of sepsis acute kidney injury are established by using the correlation data of sepsis acute kidney injury, including: S1-2-1, acquiring historical patient sign data according to the patient sign data corresponding to the correlation data of sepsis acute kidney injury; S1-2-2, acquiring historical test index data according to the test index data corresponding to the correlation data of sepsis acute kidney injury; S1-2-3, acquiring historical heart rate maximum value and historical heart rate minimum value corresponding to the historical patient sign data respectively; S1-2-4, acquiring historical arterial pressure maximum value and historical arterial pressure minimum value corresponding to the historical patient sign data respectively; S1-2-5, acquiring the difference between the historical heart rate maximum value and the historical heart rate minimum value as a historical heart rate floating value; S1-2-6, acquiring the difference between the historical arterial pressure maximum value and the historical arterial pressure minimum value as a historical arterial pressure floating value; S1-2-7, establishing a heart rate-arterial pressure interval mapping by using the historical heart rate floating value and the historical arterial pressure floating value; S1-2-8, acquiring historical test index data corresponding to the historical heart rate maximum value as heart rate peak value auxiliary data according to the historical heart rate maximum value; S1-2-9, acquiring historical test index data corresponding to the historical heart rate minimum value as heart rate valley value auxiliary data according to the historical heart rate maximum value; S1-2-10, acquiring historical test index data corresponding to the historical arterial pressure maximum value as arterial pressure peak value auxiliary data according to the historical arterial pressure maximum value; S1-2-11, acquiring historical test index data corresponding to the historical arterial pressure minimum value as arterial pressure valley value auxiliary data according to the historical arterial pressure minimum value; S1-2-12, using the heart rate-arterial pressure interval mapping, the heart rate peak value auxiliary data, the heart rate valley value auxiliary data, the arterial pressure peak value auxiliary data and the arterial pressure valley value auxiliary data as the correlation data numerical characteristics of sepsis acute kidney injury.

3. A method for sepsis acute kidney injury risk data evaluation analysis as claimed in claim 2, wherein, The correlation data trend characteristics of sepsis acute kidney injury are established by using the correlation data of sepsis acute kidney injury, including: using the time corresponding to the correlation data of sepsis acute kidney injury as a standard starting time t; acquiring patient sign data of t-1 time of the correlation data of sepsis acute kidney injury; acquiring patient sign data of t+1 time of the correlation data of sepsis acute kidney injury; acquiring test index data of t-1 time of the correlation data of sepsis acute kidney injury; The correlation data of sepsis acute kidney injury at time t+1 corresponds to the test index data; The data change trend of the test index data from time t-1 to the standard starting time t, and from the standard starting time t to time t+1 is obtained as the test index data trend; The data change trend of the test index data from time t-1 to the standard starting time t, and from the standard starting time t to time t+1 is obtained as the test index data trend; The patient sign data trend and the test index data trend are used as the correlation data trend characteristics of sepsis acute kidney injury; The data change trend is the change of adjacent data.

4. A method for sepsis acute kidney injury risk data evaluation analysis as claimed in claim 2, wherein, The risk data evaluation analysis result obtained by performing output consistency verification processing on the sepsis acute kidney injury data risk analysis model includes: S3-1, obtaining the basic data risk analysis result of sepsis acute kidney injury by using the sepsis acute kidney injury data risk analysis model; S3-2, obtaining the risk data evaluation analysis result by using the basic data risk analysis result of sepsis acute kidney injury to perform output consistency verification processing.

5. A method for sepsis acute kidney injury risk data evaluation analysis as claimed in claim 4, wherein, The basic data risk analysis result of sepsis acute kidney injury obtained by using the sepsis acute kidney injury data risk analysis model includes: The patient sign data and the test index data are input into the sepsis acute kidney injury data risk analysis model to obtain real-time first patient sign data correlation result, real-time second patient sign data correlation result, real-time first test index data correlation result, and real-time second test index data correlation result; According to the real-time first patient sign data correlation result and the real-time second patient sign data correlation result, the test index data trend corresponding to the historical first patient sign data correlation result and the historical second patient sign data correlation result is obtained as the to-be-analyzed test index data trend; According to the real-time first test index data correlation result and the real-time second test index data correlation result, the patient sign data trend corresponding to the historical first test index data correlation result and the historical second test index data correlation result is obtained as the to-be-analyzed patient sign data trend; The to-be-analyzed test index data trend and the to-be-analyzed patient sign data trend are used as the basic data risk analysis result of sepsis acute kidney injury.

6. A method for sepsis acute kidney injury risk data evaluation analysis as claimed in claim 5 wherein, The risk data evaluation analysis result obtained by using the basic data risk analysis result of sepsis acute kidney injury to perform output consistency verification processing includes: S3-2-1, judging whether the to-be-analyzed test index data trend and the real-time first test index data correlation result corresponding to the basic data risk analysis result of sepsis acute kidney injury are trend consistent, if yes, executing S3-2-2, otherwise, outputting the test index data corresponding to the real-time first test index data correlation result as the risk data evaluation analysis result; S3-2-2, judging whether the trend of the basis data risk analysis result of the sepsis acute kidney injury corresponds to the trend of the real-time second test index data, if yes, executing S3-2-3, otherwise, outputting the test index data corresponding to the real-time second test index data as the risk data evaluation analysis result; S3-2-3, judging whether the trend of the basis data risk analysis result of the sepsis acute kidney injury corresponds to the trend of the real-time first patient sign data, if yes, executing S3-2-4, otherwise, outputting the patient sign data corresponding to the real-time first patient sign data as the risk data evaluation analysis result; S3-2-4, judging whether the trend of the basis data risk analysis result of the sepsis acute kidney injury corresponds to the trend of the real-time first patient sign data, if yes, updating the correlation data of the sepsis acute kidney injury and returning to S1-2, otherwise, outputting the patient sign data corresponding to the real-time second patient sign data as the risk data evaluation analysis result; Wherein, the trend consistency is that the change of adjacent data is the same.

Images