Critical patient death risk prediction method based on big data

Abstract

The invention relates to the technical field of analysis and prediction of medical data, and discloses a critical patient death risk prediction method based on big data, which comprises the following steps of: acquiring physiological data of a human body in a normal state, establishing a standard physiological data set, constructing an analysis model based on the standard data set and in combination with a Wi-ener process, and predicting the death risk of a critical patient. The reasonable distribution range of the physiological data under different conditions is predicted by using the analysis model; then constructing an LSTM neural network model and outputting abnormal real-time body data and critical diseases related to the abnormal real-time body data; and inputting the screened abnormal real-time body data into an analysis model, screening related critical diseases, evaluating the death risk of critical patients, and providing a more intuitive and reliable prediction basis for medical personnel.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data analysis, and in particular to a method for predicting the risk of death of critically ill patients based on big data. Background Art

[0002] At present, the prediction of death risk for critically ill patients mainly relies on the experience and subjective judgment of medical staff. This method of predicting death risk lacks objective support, and there are many uncertainties in the prediction results, which cannot meet the accuracy and objectivity requirements of predicting death risk for critically ill patients.

[0003] In order to solve the above technical problems, a person skilled in the art has a patent for a system for predicting the risk of death of a patient. The publication number of the patent is: This patent uses a data acquisition module to obtain data such as the patient's medical history, the patient's physiological parameters, and the patient's laboratory parameters, wherein the physiological parameters include the patient's Glasgow score, the patient's heart rate, the patient's systolic blood pressure, and the patient's oxygen saturation; and the laboratory parameters include the patient's hemoglobin level, white blood cell count, creatinine concentration, blood potassium concentration, blood sodium concentration, urea content, platelet count, total bilirubin concentration, D-dimer level, and fibrinogen content; and then the module for calculating the risk of death of the patient is used to calculate based on the above data, thereby calculating the patient's 7-day mortality rate p. This system can effectively improve the accuracy of the prediction of the risk of death.

[0004] However, in the process of implementing the death risk prediction in the above patent, a large amount of data needs to be collected and analyzed. These main sentences will have various changes according to different patients and different states of patients. This leads to the need for a large amount of data collection and calculation in the entire death risk prediction process, which not only increases the difficulty of prediction, but also increases the calculation cost of death prediction. Therefore, those skilled in the art need a more effective method for predicting the death risk of critically ill patients with lower calculation difficulty and cost. Summary of the invention

[0005] The purpose of the present invention is to solve the above problems and to design a method for predicting the risk of death of critically ill patients based on big data.

[0006] The technical solution of the present invention to achieve the above-mentioned purpose is a method for predicting the risk of death of critically ill patients based on big data, which includes the following contents:

[0007] Establishing a standard data set: collecting physiological data of the human body in a normal state, using the collected physiological data of the human body in a normal state as standard data, and forming a standard physiological data set;

[0008] Physiological data include, but are not limited to, age, heart rate, systolic blood pressure, coma index, oxygenation index, pH, temperature, white blood cells, bilirubin, creatinine, mechanical ventilation or platelets and other parameters.

[0009] Data preprocessing: Use appropriate data processing methods to preprocess the data in the standard physiological data set, including missing value filling and data normalization, to obtain input variables. The input variables are collected by taking the median, maximum and minimum values ​​of each hour to improve the accuracy and representativeness of the data;

[0010] It should be noted that data processing methods include but are not limited to: binning method, regression method and clustering method.

[0011] Constructing an analysis model: constructing an analysis model based on a standard data set and combining it with the Wiener process, and then using the analysis model to predict the reasonable distribution range of physiological data under different conditions;

[0012] Construct a death risk prediction model: Construct an LSTM neural network model, input the real-time physical data of critically ill patients as input variables into the model, screen the real-time physical data of critically ill patients and output abnormal real-time physical data and critical diseases related to the abnormal real-time physical data; the LSTM neural network model includes two LSTM layers, the first LSTM layer includes 32 neurons, and the second LSTM layer includes 16 neurons. During the model training process, select the Adam optimization algorithm to improve the screening accuracy of the model.

[0013] Death risk assessment: The screened abnormal real-time physical data are input into the analysis model. The analysis model adjusts the reasonable distribution range according to the type of real-time physical data input, and screens related critical diseases. At the same time, the analysis model analyzes the abnormal real-time physical data and determines the extent to which the abnormal real-time physical data exceeds the reasonable distribution range. Then, the death risk of critically ill patients is assessed based on the extent to which the abnormal real-time physical data exceeds the reasonable distribution range and the types of critical diseases that have been screened, providing medical staff with a more intuitive and reliable prediction basis.

[0014] It should be noted that the construction process of the analysis model is:

[0015] Build a data analysis model for a one-dimensional linear Wiener process:

[0016] X(t)=μt+σB(t) (1)

[0017] Where, t is the time when physiological data is generated; X(t) is the reasonable distribution level of physiological data at time t; μ is the drift parameter; σ is the diffusion parameter; B(t) is the standard Brownian motion;

[0018] Then the maximum likelihood function of the segment distribution degree is:

[0019]

[0020] Among them, x ij represents the distribution level of the i-th sample at the j-th cycle; i = 1, 2, ... n; j = 1, 2, ... m; Δt is the time difference;

[0021] The maximum likelihood estimates of μ and σ are:

[0022]

[0023] X im is the distribution level of the i-th sample in the m-th cycle.

[0024] 5. According to a method for predicting the risk of death of critically ill patients based on big data according to claim 1, it is characterized in that the LSTM neural network model includes:

[0025] The input layer is used to receive real-time physical data of critically ill patients in a time series format;

[0026] The LSTM layer, as the core part of the network, is used to capture the long-term dependencies in time series data. The LSTM neural network model includes two LSTM layers. The first LSTM layer includes 32 neurons, and the second LSTM layer includes 16 neurons. During the model training process, the Adam optimization algorithm is selected as the screening method for the LSTM layer.

[0027] Fully connected layer: located after the LSTM layer, used to process the extracted features;

[0028] Output layer: An activation function is selected to generate prediction results, where the activation function includes but is not limited to: si gmo id activation function and softmax activation function.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] 1. The present invention integrates and analyzes big data from multiple medical information systems, including but not limited to basic patient information, medical history records, laboratory test results, imaging data, physiological parameter monitoring data, etc., to more comprehensively capture the key factors that affect the progression of critically ill patients, thereby significantly improving the accuracy of predicting the risk of death; this not only helps doctors identify high-risk patients earlier, but also provides a scientific basis for formulating more accurate and effective treatment plans.

[0031] 2. The accurate death risk prediction of the present invention can help medical institutions to allocate resources rationally, such as allocating limited medical resources (such as advanced life support equipment and professional medical staff) to patients who need them most, reducing resource waste and improving overall treatment efficiency and quality. This is of great significance for alleviating the shortage of medical resources and improving the accessibility of medical services.

[0032] 3. Based on the analysis results of big data, the present invention can tailor a treatment plan for each critically ill patient, taking into account individual differences and achieving more accurate treatment intervention. This helps to reduce unnecessary medical interventions, reduce the burden on patients, and improve treatment effects and patient satisfaction.

[0033] 4. The present invention predicts the risk of death of critically ill patients by abnormal data analysis and comparison, without the need for large amounts of data calculations, thereby effectively reducing the difficulty and cost of predicting the risk of death and improving medical efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a flow chart of a method for predicting the risk of death of critically ill patients based on big data according to the present invention;

[0035] Figure 2 A data table showing the reasonable distribution range of physiological data predicted by the analysis model of the present invention under different conditions;

[0036] Figure 3 It is a comparison table of the relationship between abnormal real-time physical data and critical diseases described in the present invention;

[0037] Figure 4 This is the LSTM neural network model diagram described in the present invention. DETAILED DESCRIPTION

[0038] The present invention will be described in detail below in conjunction with the accompanying drawings. Figure 1 As shown;

[0039] Embodiment 1;

[0040] Create a standard dataset:

[0041] Collect physiological data of the human body in normal state, including but not limited to: age, heart rate, systolic blood pressure, coma index, oxygenation index, pH, temperature, white blood cells, bilirubin, creatinine, mechanical ventilation status or platelet count and other parameters.

[0042] The collected physiological data is used as standard data to form a standard physiological data set, providing a benchmark for subsequent data analysis and model building.

[0043] Data preprocessing:

[0044] The data processing module is used to preprocess the data in the standard physiological data set, including missing value filling and data normalization.

[0045] The data collection method is to take the median, maximum and minimum values ​​every hour to improve the accuracy and representativeness of the data.

[0046] Build the analysis model:

[0047] An analysis model is constructed based on standard data sets and combined with the Wiener process to predict the reasonable distribution range of physiological data under different conditions.

[0048] The analytical model is used to further refine the reasonable distribution range of physiological data and provide a basis for subsequent risk assessment.

[0049] The construction process of the analysis model is as follows:

[0050] Build a data analysis model for a one-dimensional linear Wiener process:

[0051] X(t)=μt+σB(t) (1)

[0052] Where t is the time when physiological data is generated; X(t) is the reasonable distribution level at time t; μ is the drift parameter; σ is the diffusion parameter; B(t) is the standard Brownian motion;

[0053] Then the maximum likelihood function of the segment distribution degree is:

[0054]

[0055] Among them, x ij represents the distribution level of the i-th sample at the j-th cycle; i = 1, 2, ... n; j = 1, 2, ... m; Δt is the time difference;

[0056] The maximum likelihood estimates of μ and σ are:

[0057]

[0058] X im is the distribution level of the i-th sample in the m-th cycle.

[0059] Constructing a mortality risk prediction model:

[0060] Construct an LSTM neural network model, including two LSTM layers, where the first LSTM layer contains 32 neurons and the second LSTM layer contains 16 neurons.

[0061] The real-time physical data of critically ill patients are input into the model as input variables. The model screens the input data and outputs abnormal real-time physical data and critical diseases related to the abnormal data.

[0062] During the model training process, the Adam optimization algorithm was selected to improve the screening accuracy of the model.

[0063] Among them, the LSTM neural network model includes:

[0064] The input layer is used to receive real-time physical data of critically ill patients in a time series format;

[0065] The LSTM layer, as the core part of the network, is used to capture the long-term dependencies in time series data. The LSTM neural network model includes two LSTM layers. The first LSTM layer includes 32 neurons, and the second LSTM layer includes 16 neurons. During the model training process, the Adam optimization algorithm is selected as the screening method for the LSTM layer.

[0066] Fully connected layer: located after the LSTM layer, used to process the extracted features;

[0067] Output layer: An activation function is selected to generate prediction results, where the activation function includes but is not limited to: sigmoid activation function and softmax activation function.

[0068] Mortality risk assessment:

[0069] The screened abnormal real-time physical data are input into the analysis model, and the model adjusts the reasonable distribution range according to the type of the input real-time physical data.

[0070] Screen for relevant critical illnesses and analyze the extent to which abnormal real-time physical data exceeds the reasonable distribution range.

[0071] Based on the degree of abnormal data and the types of critical diseases screened out, the death risk of critically ill patients is assessed, providing medical staff with an intuitive prediction basis.

[0072] During the implementation of the present application, data collection is carried out through the hospital's information system or wearable devices to collect real-time physiological data of critically ill patients, including but not limited to the parameters listed above; and data preprocessing is to use data processing libraries (such as pandas, numpy, etc.) in programming languages ​​such as Python to preprocess the collected data, including filling missing values ​​and normalization.

[0073] It should be noted that the construction of the analysis model is to use machine learning frameworks (such as TensorFlow or PyTorch) to build a Wiener process analysis model and an LSTM neural network model. The Wiener process model is used to predict the reasonable distribution range of physiological data, and the LSTM model is used for real-time data analysis and mortality risk prediction.

[0074] Among them, model training is to use standard data sets to train the LSTM model, and select the Adam optimization algorithm to update parameters to improve the prediction accuracy of the model.

[0075] In the technical solution of the present invention, risk assessment is to input the physiological data of critically ill patients collected in real time into the trained LSTM model, and the model outputs abnormal data and related diseases. Then, the abnormal data is input into the Wiener process analysis model, and the risk of death is assessed according to the degree of excess of the data and the type of disease. The assessment results are output to medical staff in the form of charts or reports, providing an intuitive and reliable basis for their clinical decision-making.

[0076] The above technical solutions only reflect the preferred technical solutions of the technical solutions of the present invention. Some changes that may be made to certain parts thereof by technicians in this technical field all reflect the principles of the present invention and fall within the protection scope of the present invention.

Claims

1. A method for predicting the risk of death of critically ill patients based on big data, characterized in that: The method includes the following: Establishing a standard data set: collecting physiological data of the human body in a normal state, and using the collected physiological data as standard data to form a standard physiological data set; Data preprocessing: preprocess the data in the standard physiological data set to remove invalid and abnormal data; Constructing an analytical model: constructing an analytical model based on a standard data set and combining it with the Wiener process, and then using the analytical model to predict the reasonable distribution range of physiological data under different conditions; Construct a death risk prediction model: Construct an LSTM neural network model, input the real-time physical data of critically ill patients into the model as input variables, screen the real-time physical data of critically ill patients and output abnormal real-time physical data and critical diseases related to the abnormal real-time physical data; Mortality risk assessment: The screened abnormal real-time physical data are input into the analysis model. The analysis model adjusts the reasonable distribution range according to the type of real-time physical data input. At the same time, the relevant critical diseases are screened and the degree to which the abnormal real-time physical data exceeds the reasonable distribution range is determined. Then, the death risk of critically ill patients is assessed based on the degree to which the abnormal real-time physical data exceeds the reasonable distribution range and the types of critical diseases that have been screened, providing medical staff with a more intuitive and reliable prediction basis.

2. A method for predicting the risk of death of critically ill patients based on big data according to claim 1, characterized in that: The physiological data include, but are not limited to, age, heart rate, systolic blood pressure, coma index, oxygenation index, pH, temperature, white blood cells, bilirubin, creatinine and platelet parameters.

3. The method for predicting the risk of death of critically ill patients based on big data according to claim 1, characterized in that: The data preprocessing method includes: filling missing values ​​and normalizing data to obtain input variables; the input variables are collected by taking the median, maximum value and minimum value of each hour.

4. The method for predicting the risk of death of critically ill patients based on big data according to claim 1, characterized in that: The construction process of the analysis model is as follows: Build a data analysis model for a one-dimensional linear Wiener process: X(t)=μt+σB(t) (1) Where, t is the time when physiological data is generated; X(t) is the reasonable distribution level of physiological data at time t; μ is the drift parameter; σ is the diffusion parameter; B(t) is the standard Brownian motion; Then the maximum likelihood function of the segment distribution degree is: Among them, x ij represents the distribution level of the i-th sample at the j-th cycle; i = 1, 2, ... n; j = 1, 2, ... m; Δt is the time difference; The maximum likelihood estimates of μ and σ are: X im is the distribution level of the i-th sample in the m-th cycle.

5. The method for predicting the risk of death of critically ill patients based on big data according to claim 1, characterized in that: The LSTM neural network model includes: The input layer is used to receive real-time physical data of critically ill patients in a time series format; The LSTM layer, as the core part of the network, is used to capture the long-term dependencies in time series data. The LSTM neural network model includes two LSTM layers. The first LSTM layer includes 32 neurons, and the second LSTM layer includes 16 neurons. During the model training process, the Adam optimization algorithm is selected as the screening method for the LSTM layer. Fully connected layer: located after the LSTM layer, used to process the extracted features; Output layer: An activation function is selected to generate prediction results, where the activation function includes but is not limited to: si gmo id activation function and softmax activation function.

Images