Time sequence-based blood transfusion prediction method for postoperative blood condition of patient

Abstract

The invention provides a time sequence-based blood transfusion prediction method for postoperative blood conditions of patients, which belongs to the technical field of medical treatment, and comprises the following steps of: firstly, collecting basic information, past history, operation diagnosis information, blood information and curve information of a plurality of patients, screening out patients with normal data, and determining whether the patients are normal or not; the method comprises the following steps: firstly, fitting a postoperative hb recovery curve model of normal people, and then, combining the basic information and blood information of the patient collected before after each patient; according to the method, the evaluation of the postoperative condition of the patient is more targeted and accurate through the calculation of the deviation degree, a doctor can be helped to more accurately judge the postoperative condition of the patient, the risk of misdiagnosis or missed diagnosis is reduced, and more information can be provided to help the doctor to formulate a personalized treatment scheme; and meanwhile, manual participation is effectively reduced, the working efficiency is improved, the concept of the deviation degree also provides a new research view angle for researchers, and the method has wide application prospects.

Description

Technical Field

[0001] The present invention belongs to the field of medical technology, and in particular relates to a blood transfusion prediction method based on a time series of a patient's postoperative blood condition. Background Art

[0002] Medical surgeries, especially major surgeries, can cause bleeding. Patients' hemoglobin (Hb) is usually lower after surgery than before surgery. For most patients, Hb will gradually recover after surgery. However, some patients may have abnormal Hb recovery due to various factors before, during, and after surgery, and may even require postoperative blood transfusions. Early detection of these patients with abnormal Hb recovery after surgery and early intervention may reduce allogeneic blood transfusions and postoperative complications.

[0003] In the past, the clinic only used the method of closely monitoring Hb to determine whether the patient had postoperative bleeding and whether a blood transfusion was needed. However, there are controversies about whether the single Hb monitoring, the reasonableness of the blood drawing interval, the accuracy of the prediction of continued bleeding, and whether the blood transfusion decision takes into account other comorbidities of the patient. Therefore, we proposed a blood transfusion prediction method based on the time series of the patient's postoperative blood status to solve the above problems. Summary of the invention

[0004] The purpose of the present invention is to provide a blood transfusion prediction method based on the time series of the patient's postoperative blood condition, aiming to solve the problem that in the prior art, only the method of closely monitoring Hb is used to determine whether the patient has postoperative bleeding and whether a blood transfusion is needed. However, the single Hb monitoring cannot determine whether the blood drawing interval is reasonable, whether the prediction of continued bleeding is accurate, and whether the blood transfusion decision takes into account other comorbidities of the patient.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A blood transfusion prediction method based on time series of a patient's postoperative blood condition comprises the following steps:

[0007] S1. First, collect the basic information, medical history, surgical diagnosis information, blood information and curve information of multiple patients, screen out patients with normal data, and fit the postoperative hb recovery curve model of the normal population;

[0008] S2. Secondly, after each patient's operation, the patient's postoperative Hb recovery curve is predicted based on the patient's preoperative Hb and Hb on the day of operation, combined with the patient's basic information and blood information collected previously. Then, based on the deviation between the predicted Hb recovery curve and the recovery curve of the normal population, the deviation between the patient and the normal population is calculated.

[0009] S3. Finally, the best deviation classification threshold is obtained, and a decision is made on whether postoperative blood transfusion is needed or the effect of postoperative blood transfusion is evaluated based on the comparison between the current patient deviation and the best threshold.

[0010] As a preferred solution of the present invention, in step S1, the basic information of the patient includes the patient's name, age and weight.

[0011] As a preferred solution of the present invention, in step S1, the blood information includes the preoperative Hb value, the preoperative blood transfusion volume, the intraoperative blood transfusion volume, the postoperative Hb value 0-7 days and the preoperative test.

[0012] As a preferred solution of the present invention, in step S1, the curve information includes the Hb slope, the predicted curve, the lowest point and the relative change with the lowest point on the day before the operation.

[0013] As a preferred solution of the present invention, in step S1, the normal patient is a surgical patient who undergoes intraoperative blood transfusion but not postoperative blood transfusion.

[0014] As a preferred solution of the present invention, in step S1, the postoperative hb recovery curve of the normal population is fitted by LSTM and artificial intelligence model modeling.

[0015] As a preferred solution of the present invention, the artificial intelligence model is a time series prediction model, which is used to model and predict time series data, and includes the following steps:

[0016] A: Data collection, first, collect Hb measurement data of normal people, and the measurement data are arranged in time series;

[0017] B: Data preprocessing: Check if there are any outliers or missing values ​​in the data and process them;

[0018] C: Model identification: Determine the order of autoregressive (AR) and moving average (MA), where:

[0019] The autoregressive (AR) order is the number of past values ​​that need to be considered to predict the current value.

[0020] The moving average (MA) order is the number of lags of the error term that needs to be considered to predict the current error term;

[0021] D: Parameter estimation: Use maximum likelihood to estimate the parameters of the ARMA model;

[0022] E: Model diagnostics: Check the model's fit and ensure that the residuals are white noise;

[0023] F: Prediction: Use the identified and parameter estimated ARMA model to fit the Hb values ​​of the normal population.

[0024] As a preferred solution of the present invention, the time series prediction model is calculated as follows:

[0025]

[0026] Where:

[0027] Y t is the Hb value at time t;

[0028] c is a constant term, representing the Hb level without autoregressive or moving average effects;

[0029] is the autoregressive coefficient, which indicates the correlation between the current Hb value and the past i Hb values;

[0030] ∈ t is the error term, which represents the random variation of the current Hb value;

[0031] θ j is the moving average coefficient, which represents the correlation between the current error term and the past j error terms.

[0032] As a preferred solution of the present invention, the deviation calculation in step S3 uses the DTW distance based on the sum of Euclidean distances as the deviation.

[0033] As a preferred solution of the present invention, the DTW time series distance calculation process is as follows:

[0034] For two given time series Q = {q_1, q2, ..., qn) C = {c_1, c_2, ..., c_m), the DTW algorithm will first calculate the Euclidean distances between the sequence elements to obtain an nxm distance matrix D, and then search for the shortest distance from d 11 to d nm through a dynamic programming algorithm. The reciprocal of the shortest distance is the deviation between the time series Q and C.

[0035] The deviation is calculated using the DTW time series distance calculation method, and a decision on whether postoperative blood transfusion is needed is determined based on the value of the deviation.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. In the present invention, a new method for prediction and calculation is provided based on a large amount of normal patient information, such as gender, age, weight, preoperative Hb value, preoperative blood transfusion volume, intraoperative blood transfusion volume, preoperative complications, etc. Through this method, it is hoped that the postoperative Hb recovery status and the decision on whether to transfuse blood after surgery can be evaluated more accurately and automatically. By referring to the new evaluation indicators, it is possible to optimize the blood transfusion treatment of patients, thereby reducing the waste of blood resources and transfusion complications, and promoting the improvement of medical quality.

[0038] 2. In the present invention, the calculation of the deviation makes the evaluation of the patient's postoperative condition more targeted and accurate. It can not only help doctors judge the patient's postoperative condition more accurately and reduce the risk of misdiagnosis or missed diagnosis, but also provide more information to help doctors formulate personalized treatment plans; at the same time, it effectively reduces manual participation and improves work efficiency. The concept of deviation also provides researchers with a new research perspective and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0040] Figure 1 A flowchart of a method for predicting blood transfusion based on time series of a patient's postoperative blood condition according to the present invention;

[0041] Figure 2 A modeling diagram of the time series prediction model of the present invention;

[0042] Figure 3 This is a comparison chart of Hb trends after surgery in normal and abnormal patients in the present invention.

[0043] Figure 4 It is the DTW distance threshold diagram in the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] See also Figure 1 , the present invention provides the following technical solutions:

[0046] A blood transfusion prediction method based on time series of a patient's postoperative blood condition comprises the following steps:

[0047] S1. First, collect the basic information, medical history, surgical diagnosis information, blood information and curve information of multiple patients, screen out patients with normal data, and fit the postoperative hb recovery curve model of the normal population;

[0048] S2. Secondly, after each patient's operation, the patient's postoperative Hb recovery curve is predicted based on the patient's preoperative Hb and Hb on the day of operation, combined with the patient's basic information and blood information collected previously. Then, based on the deviation between the predicted Hb recovery curve and the recovery curve of the normal population, the deviation between the patient and the normal population is calculated.

[0049] S3. Finally, the best deviation classification threshold is obtained, and a decision is made on whether postoperative blood transfusion is needed or the effect of postoperative blood transfusion is evaluated based on the comparison between the current patient deviation and the best threshold.

[0050] In a specific embodiment of the present invention, the present invention is used to evaluate the postoperative blood recovery of patients. The core theme is to fit the postoperative hemoglobin (Hb) recovery curve of normal people and calculate the deviation between patients and normal people. The deviation can be used as a new postoperative recovery evaluation indicator. The present invention will screen out normal people and establish a hemoglobin recovery curve for normal people based on the postoperative recovery of normal people. Then, the postoperative hemoglobin recovery curve of patients is predicted by the model. The degree of separation between the normal curve and the predicted curve, that is, the deviation, can be used to evaluate the postoperative condition of patients. In the present invention, the calculation of the deviation makes the evaluation of the postoperative condition of patients more targeted and accurate. It can not only help doctors judge the postoperative condition of patients more accurately and reduce the risk of misdiagnosis or missed diagnosis, but also provide more information to help doctors formulate personalized treatment plans. At the same time, it effectively reduces manual participation and improves work efficiency. The concept of deviation also provides researchers with a new research perspective and has broad application prospects.

[0051] Specifically, in the present invention, it is first necessary to collect enough patient information, including basic information, past history, surgical diagnosis information, blood information and curve information, among which the basic information includes the patient's gender, age, weight, etc.; the past history should include complications such as coronary heart disease, because the transfusion indications of patients with coronary heart disease are more positive; surgical diagnosis information is also extremely important for patient transfusion evaluation and prediction, and different perioperative bleeding and transfusion situations are very different; blood information includes preoperative Hb value, preoperative transfusion volume, intraoperative transfusion volume (red blood cells, platelets, plasma), Hb value 0-7 days after surgery, preoperative tests, etc.; curve information includes the Hb slope on the day of the preoperative surgery, the predicted curve and the lowest point, and the relative change with the lowest point. Then, the data of normal patients are screened. Normal patients are defined as surgical patients who have intraoperative blood transfusion but not postoperative blood transfusion (excluding some patients with special blood diseases and critical conditions); finally, the postoperative hb recovery curve of the normal population is fitted through artificial intelligence models such as LSTM; secondly, for each patient after surgery, the trained artificial intelligence model is used to predict the postoperative hb recovery curve of the patient based on the basic information and blood information of the patient before transfusion. Then, the postoperative Hb recovery curve of the patient is predicted based on the patient's preoperative Hb and Hb on the day of surgery. Next, according to the deviation between the predicted curve and the recovery curve of the normal population, the deviation between the patient and the normal population is calculated; finally, the training set and the test set are split, and the optimal deviation classification threshold is obtained through different combinations to maximize the classification accuracy. According to the comparison between the current patient deviation and the optimal threshold, a decision is made on whether postoperative blood transfusion is needed, or the effect of postoperative blood transfusion is evaluated.

[0052] Specifically, the artificial intelligence model is a time series prediction model, which is used to model and predict time series data, including the following steps:

[0053] A: Data collection, first, collect Hb measurement data of normal people, and the measurement data are arranged in time series;

[0054] B: Data preprocessing: Check if there are any outliers or missing values ​​in the data and handle them appropriately. If there are obvious trends or seasonal components in the data, these components need to be removed from the data first so that the ARMA model can better fit the random part of the data;

[0055] C: Model identification: Determine the order of autoregressive (AR) and moving average (MA), where:

[0056] The autoregressive (AR) order is the number of past values ​​that need to be considered to predict the current value.

[0057] The moving average (MA) order is the number of lags of the error term that needs to be considered to predict the current error term;

[0058] Usually involves analyzing the autocorrelation function (ACF) and partial autocorrelation function (PACF). ACF is used to view the correlation between the current value and its previous series of values, while PACF is used to view the correlation between the current value and one or several specific values ​​before it. By observing the decline rate of ACF and PACF, the order of AR and MA can be determined;

[0059] D: Parameter estimation: Use maximum likelihood estimation or other appropriate statistical methods to estimate the parameters of the ARMA model. These parameters include the coefficients of the AR term, the coefficients of the MA term, and the variance of the error term;

[0060] E: Model diagnostics: Check the fit of the model and make sure the residuals are white noise. This can be done by plotting the ACF and PACF of the residual sequence. If the residual sequence shows any pattern, the order of the model may need to be reconsidered.

[0061] F: Prediction: Use the identified and parameter-estimated ARMA model to predict the future Hb values ​​of the normal population. You can choose different prediction periods, for example, predict the Hb levels in the next few days or a week.

[0062] The time series forecasting model is calculated as follows:

[0063]

[0064] Where:

[0065] Y t is the Hb value at time t;

[0066] c is a constant term, representing the Hb level without autoregressive or moving average effects;

[0067] is the autoregressive coefficient, which indicates the correlation between the current Hb value and the past i Hb values;

[0068] ∈ t is the error term, which represents the random variation of the current Hb value;

[0069] θ j is the moving average coefficient, which represents the correlation between the current error term and the past j error terms;

[0070] When calculating, you need to use statistical software, such as R, Python's Pandas and Statsmodels libraries, MATLAB and other software to perform ARMA model identification, parameter estimation and prediction. By comparing the predicted values ​​with the actual values, the prediction performance of the ARMA model can be evaluated. Different statistical indicators such as mean square error (MSE), root mean square error (RMSE) or mean absolute error (MAE) can be used to evaluate the accuracy of the model.

[0071] Specifically, the deviation calculation uses the DTW distance based on the sum of the Euclidean distance as the deviation, and both the normal population curve and the patient curve are represented as time series data, where each data point contains the Hb value at the corresponding time. The DTW time series distance is calculated for the Euclidean distance matrix of all time point pairs to obtain the total deviation of the patient curve relative to the normal population curve; the calculation process of the DTW time series distance is as follows:

[0072] For two given time series Q = {q_1, q2, …, qn) C = {c_1, c_2, …, c_m), the DTW algorithm will first calculate the Euclidean distances between the sequence elements in turn to obtain an nxm distance matrix D, and then search for the shortest distance from d 11 to d nm through a dynamic programming algorithm. The reciprocal of the shortest distance is the deviation between the time series Q and C.

[0073] The implementation of the present invention specifically comprises the following steps:

[0074] First, it is necessary to collect enough patient information through appropriate means. Preferably, an appropriate database and data collection system deployed on the hospital's internal server should be established to collect the basic information of the patient before the operation, test information, diagnosis and nursing text information, and the hb test value at least 7 days after the operation;

[0075] Then, the data needs to be properly cleaned and processed to meet the needs of modeling. The data obtained in the previous step is supplemented with missing values, irrelevant features are deleted, and features are added according to the business.

[0076] Next, it is necessary to fit the Hb recovery curve of the normal population through a suitable method. The main model selection range is the model that can process time series data, such as ARMA. ARMA is a time series prediction model used to model and predict time series data. The ARMA model combines the characteristics of autoregression (AR) and moving average (MA). The AR part considers the impact of the value of the past time point on the current value, while the MA part considers the impact of the past error on the current value. The characteristic of ARMA is that it can adapt to different time series patterns and has good prediction ability.

[0077] Then, the normal population curve model that has been fitted, as well as information such as the patient's preoperative Hb and Hb on the day of surgery, is used to predict the patient's postoperative Hb recovery curve, and the deviation between the patient curve and the normal population curve is calculated. The patient curve and the normal population curve can be regarded as two sets of hb time series data, and the DTW time series distance is used as the deviation.

[0078] Finally, based on the degree of deviation, a decision is made as to whether postoperative blood transfusion is needed. The normal population curve, the patient's curve, and the classification threshold are displayed on the appropriate page to prompt the doctor of the patient's postoperative recovery status and assist the doctor in making clinical decisions.

[0079] In the present invention, a new method for prediction and calculation is provided based on a large amount of normal patient information, such as gender, age, weight, preoperative Hb value, preoperative blood transfusion volume, intraoperative blood transfusion volume, preoperative complications, etc. Through this method, it is hoped that the postoperative Hb recovery status and the decision on whether to transfuse blood after surgery can be evaluated more accurately and automatically. By referring to the new evaluation indicators, it is possible to optimize the blood transfusion treatment of patients, thereby reducing the waste of blood resources and transfusion complications, and promoting the improvement of medical quality.

[0080] Example 2

[0081] This embodiment 2 also provides a method for calculating a reasonable Hb recovery curve after surgery for a blood transfusion prediction method for a patient's postoperative blood condition based on a time series, which is as follows:

[0082] Specifically, LSTM (Long Short-Term Memory) is a special recursive neural network (RNN) designed to solve the gradient vanishing and gradient exploding problems of standard RNN when processing long sequence data. LSTM can better capture and remember long-term dependencies by introducing a gating mechanism. The training set of LSTM only considers normal patients, so after the modeling is completed, the model can be used to predict the more reasonable recovery curve of patients after surgery.

[0083] Specifically, Figure 2 As shown in the figure, the actual postoperative Hb recovery curve of the patient and the normal population curve can be regarded as two sets of Hb time series data. Then, the common DTW time series distance calculation method is used to calculate the deviation. According to the deviation value, a decision can be made whether postoperative blood transfusion is needed. In actual application, the normal curve and the patient's predicted curve and classification threshold can be displayed on a suitable page. When the deviation between the patient's predicted Hb recovery curve and the normal curve is greater than the specified threshold, the doctor can be prompted to pay attention to the patient's postoperative recovery and assist the doctor in making clinical decisions.

[0084] like Figure 3 As shown in the figure, the comparison of postoperative Hb trends between normal and abnormal patients can be clearly seen, where the horizontal axis represents the number of days after surgery and the vertical axis represents the postoperative Hb value; the box plot describes the Hb distribution of patients every day after surgery; the upper curve describes the postoperative daily Hb trend of normal patients, and the lower curve describes the postoperative daily Hb trend of abnormal patients;

[0085] Among them, DTW is used as the calculation process of the deviation index, such as Figure 4 As shown in the figure, Dynamic Time Warping (DTW) is a common method to measure the similarity between two time series. For a given two time series Q = {q_1, q2, ..., qn) C = {c_1, c_2, ..., c_m), the DTW algorithm will first calculate the Euclidean distance between the sequence elements to obtain an nxm distance matrix D, and then search for the shortest distance from d11 to dnm through a dynamic programming algorithm. The reciprocal of the shortest distance is the deviation between the time series Q and C.

[0086] The deviation threshold is set based on whether or not postoperative blood transfusion has occurred. The patient group naturally has type attributes. Therefore, the principle of maximizing classification accuracy can be used to find the optimal deviation threshold for type differentiation and use it as the initial value of the deviation. The value of the deviation threshold can be flexibly adjusted according to the actual medical scenario, but it may involve advice and communication with the doctors or medical team who hold the decision-making power.

[0087] Specifically, the calculation method of the deviation threshold is as follows:

[0088] 1. After a patient is given and sufficient historical data of the patient is collected, the postoperative Hb recovery curve of the patient under normal conditions is fitted with LSTM, i.e., the normal curve; the actual postoperative Hb recovery curve of the patient is fitted with the ARMA algorithm of a specific single disease, i.e., the predicted curve;

[0089] 2. Calculate DTW as a deviation indicator based on the patient's normal curve and predicted curve;

[0090] 3. For all historical patients, the corresponding DTW distance is calculated, and the patient population is divided into two categories according to whether postoperative bleeding occurs. The label of postoperative bleeding is 1, and the label of postoperative bleeding is 0;

[0091] 4. Select different DTW distance thresholds, and count the Label distribution in the patient group with a distance less than the DTW distance threshold and the Label distribution in the patient group with a distance greater than the DTW distance threshold, and obtain the confusion matrix, as shown in Table 1 below:

[0092] Patients with DTW less than the threshold Patients with DTW greater than the threshold Patients with Label 1 TP FN Patients whose label is 0 FP TN

[0093] Table 1:

[0094] 5. Calculate the F1 statistic based on the confusion matrix. The formula is as follows:

[0095] F1 = 2 × TP / (2 × TP + FP + FN);

[0096] 6. Traverse all possible DTW distance thresholds, use the maximum F1 statistic as the evaluation basis for "optimal classification", and find the best DTW distance threshold. The specific visualization is as follows Figure 4 As shown, the vertical dotted line represents the optimal DTW distance threshold.

[0097] Finally, it should be noted that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A blood transfusion prediction method based on time series of postoperative blood status of patients, characterized in that: The steps include: S1. First, collect the basic information, medical history, surgical diagnosis information, blood information and curve information of multiple patients, screen out patients with normal data, and fit the postoperative hb recovery curve model of the normal population; S2. Secondly, after each patient's operation, the patient's postoperative Hb recovery curve is predicted based on the patient's preoperative Hb and Hb on the day of operation, combined with the patient's basic information and blood information collected previously. Then, based on the deviation between the predicted Hb recovery curve and the recovery curve of the normal population, the deviation between the patient and the normal population is calculated. S3. Finally, the best deviation classification threshold is obtained, and a decision is made on whether postoperative blood transfusion is needed or the effect of postoperative blood transfusion is evaluated based on the comparison between the current patient deviation and the best threshold.

2. The method for predicting blood transfusion based on time series of postoperative blood status of patients according to claim 1, characterized in that: In step S1, the basic information of the patient includes the patient's name, age and weight.

3. The method for predicting blood transfusion based on the time series of the patient's postoperative blood condition according to claim 2, characterized in that: In step S1, the blood information includes the preoperative Hb value, the preoperative blood transfusion volume, the intraoperative blood transfusion volume, the postoperative Hb value 0-7 days after the operation, and the preoperative test.

4. The method for predicting blood transfusion based on the time series of the patient's postoperative blood condition according to claim 3, characterized in that: In step S1, the curve information includes the Hb slope on the day before surgery, the predicted curve, the lowest point, and the relative change to the lowest point.

5. The method for predicting blood transfusion based on time series of postoperative blood status of patients according to claim 4, characterized in that: In step S1, the normal patient is a surgical patient who undergoes intraoperative blood transfusion but not postoperative blood transfusion.

6. The method for predicting blood transfusion based on the time series of the patient's postoperative blood condition according to claim 5, characterized in that: In step S2, the postoperative HB recovery curve of the normal population is fitted through LSTM and artificial intelligence model modeling.

7. The method for predicting blood transfusion based on time series of postoperative blood status of patients according to claim 6, characterized in that: The artificial intelligence model is a time series prediction model, which is used to model and predict time series data, and includes the following steps: A: Data collection, first, collect Hb measurement data of normal people, and the measurement data are arranged in time series; B: Data preprocessing: Check if there are any outliers or missing values ​​in the data and process them; C: Model identification: Determine the order of autoregressive (AR) and moving average (MA), where: The autoregressive (AR) order is the number of past values ​​that need to be considered to predict the current value. The moving average (MA) order is the number of lags of the error term that needs to be considered to predict the current error term; D: Parameter estimation: Estimate the parameters of the ARMA model using maximum likelihood; E: Model diagnostics: Check the fit of the model and ensure that the residuals are white noise; F: Prediction: Use the identified and parameter estimated ARMA model to fit the Hb values ​​of the normal population.

8. The method for predicting blood transfusion based on time series of postoperative blood status of patients according to claim 7, characterized in that: The time series prediction model is calculated as follows: Where: Y t is the Hb value at time t; c is a constant term, representing the Hb level without autoregressive or moving average effects; is the autoregressive coefficient, which indicates the correlation between the current Hb value and the past i Hb values; ∈ t is the error term, which represents the random variation of the current Hb value; θ j is the moving average coefficient, which represents the correlation between the current error term and the past j error terms.

9. The method for predicting blood transfusion based on time series of postoperative blood status of patients according to claim 8, characterized in that: The deviation calculation in step S3 uses the DTW time series distance as the deviation.

10. The method for predicting blood transfusion based on time series of postoperative blood status of patients according to claim 9, characterized in that: The DWT time series distance summation calculation process is as follows: For two given time series Q = {q_1, q2, ..., qn) C = {c_1, c_2, ..., c_m), the DTW algorithm calculates the Euclidean distances between the sequence elements in turn to obtain an nxm distance matrix D, and then searches for the shortest distance from d 11 to d nm through a dynamic programming algorithm. The reciprocal of the shortest distance is the deviation between the time series Q and C. The deviation is calculated using the DTW time series distance calculation method, and a decision on whether postoperative blood transfusion is needed is determined based on the value of the deviation.

Images