CN118692643A - A blood management decision-making support system - Google Patents

Abstract

The present application provides a blood management decision-making support system, including a data acquisition module, a blood demand prediction module, a blood collection trend prediction module and a blood use effect evaluation module; the data acquisition module is responsible for collecting blood transfusion data and blood collection data; the blood demand prediction module is responsible for analyzing historical blood use data and predicting blood demand under different circumstances; the blood collection trend prediction module is responsible for using historical blood collection data and factors affecting blood collection to predict future blood collection trends; the blood use effect evaluation module is responsible for analyzing and evaluating the rationality and effectiveness of blood use to guide blood use decisions and improve the efficiency of blood resource use. Based on the LSTM time series prediction method, a blood demand prediction model is constructed. Through the data before blood collection, based on the ARIMA time series prediction method, a blood collection structure prediction model is built, and finally a decision-making support system that can take into account multiple functions such as blood collection, use, and prediction is constructed. The system can provide hospitals and blood stations with decision-making suggestions on blood collection, allocation, and use, guide clinical rational blood use, and achieve precise blood use management.

Description

A blood management decision-making support system

Technical Field

The present invention relates to the field of blood bank management, and in particular to a blood management decision-making support system.

Background Art

As the only source of blood, voluntary blood donation plays an important role in protecting national health and national security. However, in my country, the amount of blood collected and supplied cannot meet the normal growth demand, and blood supply shortages and "blood shortages" occur from time to time.

"Blood shortage" refers to the phenomenon of blood type imbalance or emergency, where blood collection and supply medical institutions are unable to meet the blood needs of their health care institutions. There are many reasons for the "blood shortage" phenomenon in my country, such as the public's lack of awareness of voluntary blood donation, insufficient number of voluntary blood donors, unreasonable clinical use of blood, the lack of significant results of voluntary blood donation publicity by health administrative departments and blood stations, and the lack of industry supervision. Among them, the ability of blood collection and supply institutions to collect and supply blood is one of the most important influencing factors. These are the challenges and difficulties currently facing the voluntary blood donation cause.

It is very necessary to reasonably predict the changes in blood use based on the development of medical level and historical data, and arrange the development plan and policy of the blood center based on the prediction results, and formulate an effective and reasonable blood donation recruitment plan, which not only ensures the supply while avoiding waste, but also plays an important role in the development and construction of blood collection and supply institutions. At the same time, it can avoid insufficient blood inventory in the event of blood shortage or major emergencies.

Summary of the invention

The purpose of the present invention is to solve the technical problems in the above background and propose a blood management decision-making support system, including a data acquisition module, a blood demand prediction module, a blood collection trend prediction module and a blood use effect evaluation module;

The data collection module is responsible for collecting blood transfusion data and blood collection data;

The blood demand prediction module is responsible for analyzing historical blood usage data and predicting blood demand under different circumstances;

The blood collection trend prediction module is responsible for predicting future blood collection trends using historical blood collection data and factors that affect blood collection (such as weather, demographic data, etc.);

The blood use effectiveness evaluation module is responsible for analyzing and evaluating the rationality and effectiveness of blood use in order to guide blood use decisions and improve the efficiency of blood resource utilization.

In the preferred solution, the blood demand prediction module includes an LSTM time series prediction model to predict the blood demand in the next seven days;

The data collection module collects blood demand data at this time in previous years, and the blood demand prediction module compares the predicted data with the blood demand data at this time in previous years;

When the predicted blood demand is higher than the blood demand data of previous years, it directly forms (supplements) one of the elements of the formulated blood collection plan;

When the predicted blood demand is lower than the blood demand data of previous years, the blood station needs to make further adjustments and improvements before it can be used as a factor for blood collection.

In the preferred solution, the blood collection trend prediction module includes an ARIMA prediction model to predict the collection volume of various blood types in the next seven days;

The data collection module collects the data of various blood types collected at this time in previous years, and the blood collection trend prediction module compares the predicted data with the collection data of previous years:

When the predicted blood collection volume is low, the blood collection needs to be adjusted to increase the collection volume of the missing blood type;

When the predicted blood collection volume is high, the blood collection structure quantity for the next seven days will be optimized to further refine the collection volume of each blood type.

In a preferred solution, the blood inventory management module generates a blood collection plan according to the blood collection prediction result, which specifically includes the following steps:

S1, the blood sampling trend prediction module generates the predicted blood sampling amount by comparing with the historical data;

S2. The blood demand prediction model is compared with the historical blood usage data and then corrected to form the predicted blood usage amount;

S3. When the predicted blood demand is greater than the predicted blood collection, the blood collection volume for the next seven days will be increased; when the predicted blood demand is less than the predicted blood collection, the blood collection volume for the next seven days will be reduced, and the part exceeding the predicted demand will be transferred to the ischemic and anemic areas.

The preferred solution also includes an emergency demand response module;

The emergency demand response module includes a rapid response model for rapid identification, early warning and decision-making in emergency situations;

Emergency demand response includes the following steps:

S1. Collect and organize emergency situations that have occurred in history, including blood shortages, natural disasters, and large-scale accidents, and analyze the characteristics of emergencies and response measures.

S2. Use statistical and machine learning techniques to assess the probability of different emergency situations and establish a predictive model.

S3. Develop an emergency list based on historical data and risk assessment;

S4. For each emergency situation in the list, pre-set response measures;

S5. Use real-time data monitoring and automated algorithms to quickly identify emergency situations;

S6. When an emergency is detected to be about to happen or has already happened, the system can automatically issue an early warning;

S7. Provide decision support tools to help managers make decisions quickly, including the selection and implementation of preset measures.

In the preferred solution, the LSTM model is used for prediction, which is expressed as:

Where i = 1, 2, …, T; o, h and c represent the output, hidden state and cell state of LSTM respectively, T is the length of the time series window, and _ep represents the blood consumption after integrating all months in the time series window.

In the preferred scheme, the blood collection prediction method includes the following steps: sorting out historical blood collection data, checking the stationarity of the time series, and performing a difference operation when the time series is not stationary; performing an autocorrelation test on the data to determine how many periods of past data are used in the model to predict the current value; performing a partial autocorrelation test on the data to determine how many periods of residuals in the past are used in the model to predict the current value; using ARIMA model parameters to fit the model to a portion of the data, and using the remaining data to evaluate the model; using the root mean square error to evaluate the degree of fit of the model, and using the trained ARIMA model to predict the blood collection data for the next seven days.

In the preferred solution, the blood use effect evaluation module takes the blood use volume as the research object and data set, selects the factors affecting the blood use volume prediction as the variable target, and establishes a logistics regression model;

The blood use assessment includes the following steps: first, factor analysis is performed on the prediction indicators, the data is reduced in dimension, the correlation between the data is eliminated, and the final indicators are determined; then a logistics regression model is established based on the final indicators; finally, the prediction effect of the logistics regression model is tested, and suggestions for rational blood use are made based on the actual results.

In the preferred solution, the Logistics regression model is expressed as:

Among them, _x1 is blood type A, _x2 is blood type B, _x3 is blood type AB, and _x4 is blood type O; suppose the dependent variable is a binary variable, with the value y=0 or y=1, and the m variables that affect the value of y are _{x1 x2 x3 x4} ; under the influence of m independent variables, the conditional probability of the positive result y=1 is p=p(y=1| _x1 , _x2 ,… _xm );

Perform logit transformation on the Logistics regression model expression to obtain the following linear form of the Logistics regression model:

The constant term β is the natural logarithm of the ratio of the probability of adequate blood supply to that of inadequate blood supply when all exposure factors are 0; the partial regression coefficient β _j (j = 1, 2, 3, ..., m) represents the average change in logit (p) when the jth independent variable changes by one unit under the condition that other independent variables are fixed, and it has a corresponding relationship with the odds ratio OR; when other influencing factors are the same, the natural logarithm of the odds ratio of two different exposure levels of a risk factor x _j and c ₁ and c ₀ is:

so:

OR _j = exp[β _j (c ₁ −c ₀ )]

The larger the OR, the greater the probability of a positive result, indicating that the blood supply for that blood type is sufficient.

The beneficial effects of the present invention are: after machine prediction and manual modification and improvement, the blood demand forecast in the system will greatly reduce the errors occurring in the blood demand forecast - excessive forecast amounts will lead to deviations in the collection, transportation and storage plans, resulting in increased costs and blood waste; too low forecast amounts will lead to insufficient collection and insufficient blood supply; the ARIMA model is used to predict the collection amounts of various blood types in the next seven days, which can further accurately determine the structure of blood collection - the collection amounts of various blood types, and effectively avoid blood waste; the blood usage amount is integrated and learned using logistic regression to form a blood usage effect evaluation model, which will impose more precise constraints on the blood demand forecast model, and the data-driven method is used to update the model in real time, thereby improving the accuracy of blood demand forecasting.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a diagram of blood use effect evaluation in the present invention;

FIG2 is a flow chart of data collection and processing in the present invention;

FIG3 is a LSTM model prediction flow chart in the present invention;

FIG4 is an ARIMA model prediction flow chart in the present invention;

FIG5 is a flow chart of blood sampling plan formulation in the present invention;

FIG6 is a flow chart of the Logistics regression model for evaluating the effect of blood use.

DETAILED DESCRIPTION

Example 1

As shown in Figures 1 to 6, in order to avoid seasonal and structural blood shortages, the present invention proposes a system that collects hospital blood use-related data, blood station application data, and blood station historical collection data, performs three-party data fusion, data cleaning, data normalization, and data missing supplementation. In order to prevent errors in the subsequent formulation of the collection and writing plan, this processing further strengthens the correctness and relevance of the data; the processed blood transfusion data is imported into the LSTM time series prediction model and assisted by references and expert rules to predict blood demand. Demand forecasting and quantity structure forecasting together constitute the two important basic indicators of the blood collection plan. At the same time, the weather factors that will affect blood collection are considered from another dimension and multiple aspects. The blood collection plan is formulated and supplied to the blood station. The blood station prepares and transports according to its specific plan, effectively avoiding the situation of ischemia, less blood, multiple uses of one blood, and multiple grabs of one blood. When blood is delivered to the hospital and the hospital performs surgery to rescue the patient, the system will use logistic regression and expert assistance to integrate the learning of the doctor's blood usage, and finally form a blood usage effect evaluation model to assist in understanding blood needs.

1. Data cleaning and removal:

Collect data from the hospital's HIS, LIS, blood transfusion management platform, surgical anesthesia management system, blood station application data, blood station transportation data, and historical blood collection data. Collect as much blood collection, application, transportation, usage quantity and blood structure as possible in multiple and comprehensive aspects.

After eliminating useless and erroneous data, the three data are integrated and relevant indicators are extracted for the next step of analysis.

Clean up redundant and erroneous data. Find missing data and supplement it retrospectively. Reduce the dimension of data to avoid difficulties in subsequent processing. Standardize data to further optimize its data features.

2. Blood demand prediction model:

Organize historical blood transfusion data and blood station application data

Carefully organize historical blood transfusion records and blood station application and issuance data

LSTM model training and hyperparameter adjustment.

Implementation method:

Where i=1,2,…,T,o,h,c represent the output, hidden state, and cell state of LSTM respectively, T is the length of the time series window, and _ep represents the blood consumption after integrating all months in the time series window;

Based on relevant literature and expert opinions, we used the Delphi method to synthesize the experts' opinions and reach a consensus on the importance of features. Based on relevant literature and expert opinions, we used the Delphi method to synthesize the experts' opinions and reach a consensus on the importance of features.

The mean absolute error, root mean square error, and blood effect evaluation model opinions were used to evaluate the model effect.

Use the trained model to predict blood demand for the next seven days.

The model is dynamically updated based on the subsequent blood use effect evaluation.

3. Blood sampling prediction model:

Organize historical blood collection data;

Check the stationarity of the time series, and perform a difference operation if it is not stationary;

Perform an autocorrelation (ACF) test on the data to determine how many periods of past data are used in the model to predict the current value;

Perform a partial autocorrelation (PACF) test on the data to determine how many past residuals are used in the model to predict the current value;

Choose appropriate ARIMA model parameters;

A portion of the data is used for model fitting, and the remaining data is used for model evaluation;

The root mean square error (RMSE) was used to evaluate the model fit;

Use the trained ARIMA model to predict blood collection data for the next seven days.

Specific implementation:

formula:

Y _t ＝c+φ ₁ Y _(t-1) +φ ₂ Y _(t-2) +...+φ _p Y _(tp) +θ ₁ ∈ _(t-1) +θ ₂ ∈ _{(t-2 )} +...+θ _q ∈ _(tq) +∈ _t .

Y _t represents the blood collection volume at time t that we need to predict;

Among them, _φ1 to _φp are used to describe the relationship between _Yt and the blood collection volume at the past p time nodes; _θ1 to _θq are used to describe the relationship between _Yt and the error at the past q time nodes; _∈t is the error term at time t; c is the constant term.

4. ARIMA(p,d,q)

p: parameter related to the autoregressive aspect of the model, the number of lagged observations included in the model, also known as the lag order;

d: a parameter related to the integration part of the model, the number of times the original observations are different, also called the degree of difference. It affects the amount of differences applied to the time series;

q: Parameter related to the moving average part of the model. The size of the moving average window, also known as the order of the moving average.

5.Stability test:

The cleaned historical blood collection data is subjected to ADF test. If it is not stable, d-order difference is performed to transform it into a stable series;

Determine the p and q values, and perform PACF and ACF on the historical blood sampling data after d-order difference to determine the p and q values. Where p is the maximum lag point of the PACF graph, and q is the maximum lag point of the ACF graph. Use the AIC and BIC criteria to compare the AIC and BIC of different difference orders, and take the p and q with the minimum value of the two.

6. Generate blood collection plan

a. The blood collection prediction model generates accurate blood collection prediction volume by comparing with historical data;

b. The blood demand forecast model is compared with the historical blood usage data and then revised to form the predicted blood usage volume;

c. The blood demand prediction model and the blood collection demand prediction model work together:

If the predicted blood demand is greater than the predicted blood collection, the blood collection plan for the next seven days will be increased in combination with the blood inventory at the blood station; if the predicted blood demand is less than the blood collection prediction model, the blood collection plan for the next seven days will be adjusted in combination with the blood inventory at the blood station, the blood collection volume will be reduced, and the excess blood will be transferred to other blood-deficient and anemic areas;

d. Future weather forecast: The weather in the next seven days will also interfere with the blood collection plan. After predicting the weather in the next seven days, a more accurate analysis and positioning of the blood collection time in different regions can be carried out.

7. Hospital blood use effect evaluation model:

Historical blood use data were collected from hospitals, including patient demographic information, medical history, laboratory test results, and blood transfusion records;

Preprocess the data, handle missing values, outliers, and inconsistencies, and convert the data into a format suitable for analysis;

Select patient age and gender, medical history, blood type and Rh status, previous blood transfusion history from the data. Use principal component analysis (PCA) or recursive feature elimination (RFE) to extract relevant features from the data;

Use the selected features to build a logistic regression model to predict the likelihood of blood use. Use backward elimination or stepwise regression to select the most relevant features and avoid overfitting.

Use accuracy, precision, recall, and F1 score to evaluate the performance of the logistic regression model;

A technical Delphi method was used to aggregate expert opinions and obtain consensus on feature importance;

Combine the logistic regression model with expert opinions and use Bayesian model averaging to integrate the logistic regression model with expert opinions. Use precision, recall and F1 score to evaluate the performance of the hybrid model;

Cross-validation and bootstrapping were used to validate the mixed model;

Continuously monitor and update the hybrid model.

Specific implementation:

Logistics regression research method: Taking blood usage as the research object and data set, factors affecting blood usage prediction are selected as variable targets to establish a logistics regression model. First, factor analysis is performed on the prediction indicators, dimensionality reduction is performed on the data, the correlation between the data is eliminated, and the final indicators are determined; then a logistics regression model is established based on the final indicators; finally, the prediction effect of the model is tested, and reasonable suggestions for reasonable blood use are made in combination with the actual results.

Pearson's test for correlation:

Variable description: x ₁ : blood type A, x ₂ : blood type B, x ₃ : blood type AB, x ₄ : blood type O;

Step 1: Data cleaning, assume that the dependent variable is a binary variable, with the value y=0 or y=1 (whether the blood volume is sufficient). The m variables that affect the value of y are _{x1 x2 x3 x4} . Under the influence of m independent variables, the conditional probability of a positive result, y=1, is p=p(y=1| _x1 , _x2 , ... _xm ).

The Logistics regression model can be expressed as:

Step 2: Perform logit transformation on the formula to obtain the following linear form of the logistics regression model:

The constant term β is the natural logarithm of the ratio of the probability of adequate blood supply to the probability of inadequate blood supply when all exposure factors are 0. The partial regression coefficient β _j (j = 1, 2, 3, ..., m) represents the average change in logit (p) when the jth independent variable changes by one unit under the condition that other independent variables are fixed. It has a corresponding relationship with the odds ratio OR. When other influencing factors are the same, the natural logarithm of the odds ratio of two different exposure levels of a risk factor x _j and c ₁ and c ₀ is:

so:

OR _j = exp[β _j (c ₁ −c ₀ )]

The larger the OR, the greater the probability of a positive result, indicating that the blood supply for that blood type is sufficient.

Example 2

A blood management decision-making support system, comprising a data collection module, a blood demand prediction module, a blood collection trend prediction module, a resource allocation decision module, a risk assessment and mitigation module, and a blood inventory management module;

The data collection module is responsible for collecting blood transfusion data and blood collection data;

The blood demand prediction module is responsible for analyzing historical blood usage data and predicting blood demand under different circumstances;

The blood collection trend prediction module is responsible for predicting future blood collection trends using historical blood collection data and factors that affect blood collection (such as weather, demographic data, etc.);

The resource allocation decision module is responsible for making blood resource allocation decisions and optimizing inventory distribution based on forecasts and real-time data;

The risk assessment and blood collection decision module is responsible for evaluating potential risks, such as supply shortage or surplus, and proposing corresponding mitigation measures;

The blood inventory management module is responsible for tracking and managing blood inventory, including blood type classification, quantity, expiration date, etc., and providing inventory warnings.

The preferred solution also includes an emergency demand response module, which is responsible for quickly formulating and executing response plans for emergencies or unforeseen demand surges.

The preferred solution also includes a data monitoring and reporting module, which is responsible for real-time monitoring of blood collection and blood supply data, generating reports, and providing decision support for management.

In the preferred solution, the blood demand prediction module includes an LSTM time series prediction model to predict the blood demand in the next seven days;

The data collection module collects blood demand data at this time in previous years, and the blood demand prediction module compares the predicted data with the blood demand data at this time in previous years;

When the predicted blood demand is higher than the blood demand data of previous years, it directly forms (supplements) one of the elements of the formulated blood collection plan;

When the predicted blood demand is lower than the blood demand data of previous years, the blood station needs to make further adjustments and improvements before it can be used as a factor for blood collection.

In the preferred solution, the blood collection trend prediction module includes an ARIMA prediction model to predict the collection volume of various blood types in the next seven days;

The data collection module collects the data of various blood types collected at this time in previous years, and the blood collection trend prediction module compares the predicted data with the collection data of previous years:

When the predicted blood collection volume is low, the blood collection needs to be adjusted to increase the collection volume of the missing blood type;

When the predicted blood collection volume is high, the blood collection structure quantity for the next seven days will be optimized to further refine the collection volume of each blood type.

In a preferred solution, the blood inventory management module generates a blood collection plan according to the blood collection prediction result, which specifically includes the following steps:

S1, the blood sampling trend prediction module generates the predicted blood sampling amount by comparing with the historical data;

S2. The blood demand prediction model is compared with the historical blood usage data and then corrected to form the predicted blood usage amount;

S3. When the predicted blood demand is greater than the predicted blood collection, the blood collection volume for the next seven days will be increased; when the predicted blood demand is less than the predicted blood collection, the blood collection volume for the next seven days will be reduced, and the part exceeding the predicted demand will be transferred to the ischemic and anemic areas.

In the preferred solution, after predicting the weather for the next seven days, a more accurate analysis and positioning of the blood collection time in different regions is performed.

In a preferred solution, the emergency demand response module includes a rapid response model for rapid identification, warning and decision-making in emergency situations;

Emergency demand response includes the following steps:

S1. Collect and organize emergency situations that have occurred in history, including blood shortages, natural disasters, and large-scale accidents, and analyze the characteristics of emergencies and response measures.

S2. Use statistical and machine learning techniques to assess the probability of different emergency situations and establish a predictive model.

S3. Develop an emergency list based on historical data and risk assessment;

S4. For each emergency situation in the list, pre-set response measures;

S5. Use real-time data monitoring and automated algorithms to quickly identify emergency situations;

S6. When an emergency is detected to be about to happen or has already happened, the system can automatically issue an early warning;

S7. Provide decision support tools to help managers make decisions quickly, including the selection and implementation of preset measures.

In a preferred embodiment, the list of emergency situations in step S3 includes: natural disasters, public health events, blood supply shortages, blood inventory reaching its expiration date, failures at key blood collection sites, seasonal changes in blood demand, abnormally high proportions of unqualified blood found in blood testing, shortages of key personnel, and technical or system failures.

In the preferred solution, different response levels are provided according to the probability of occurrence of different emergency situations.

The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (9)

1. A blood management decision support system, characterized by: comprising a data acquisition module, a blood demand prediction module, a blood collection trend prediction module and a blood use effect evaluation module; The data collection module is responsible for collecting blood transfusion data and blood collection data; The blood demand prediction module is responsible for analyzing historical blood usage data and predicting blood demand under different circumstances; The blood collection trend prediction module is responsible for predicting future blood collection trends using historical blood collection data and factors affecting blood collection; The blood use effectiveness evaluation module is responsible for analyzing and evaluating the rationality and effectiveness of blood use in order to guide blood use decisions and improve the efficiency of blood resource utilization. 2. A blood management decision support system according to claim 1, characterized in that: the blood demand prediction module includes an LSTM time series prediction model to predict the blood demand in the next seven days; The data collection module collects blood demand data at this time in previous years, and the blood demand prediction module compares the predicted data with the blood demand data at this time in previous years; When the predicted blood demand is higher than the blood demand data of previous years, it directly forms the elements of the blood collection plan to be formulated; When the predicted blood demand is lower than the blood demand data of previous years, the blood station needs to make further adjustments and improvements before it can be used as a factor for blood collection. 3. A blood management decision support system according to claim 1, characterized in that: the blood collection trend prediction module includes an ARIMA prediction model to predict the collection volume of various blood types in the next seven days; The data collection module collects the data of various blood types collected at this time in previous years, and the blood collection trend prediction module compares the predicted data with the collection data of previous years: When the predicted blood collection volume is low, the blood collection needs to be adjusted to increase the collection volume of the missing blood type; When the predicted blood collection volume is high, the blood collection structure quantity for the next seven days will be optimized to further refine the collection volume of each blood type. 4. The blood management decision support system according to claim 1, characterized in that: it also includes a blood inventory management module; The blood inventory management module generates a blood collection plan based on the blood collection prediction results, which specifically includes the following steps: S1, the blood sampling trend prediction module generates the predicted blood sampling amount by comparing with the historical data; S2. The blood demand prediction model is compared with the historical blood usage data and then corrected to form the predicted blood usage amount; S3. When the predicted blood demand is greater than the predicted blood collection, the blood collection volume for the next seven days will be increased; when the predicted blood demand is less than the predicted blood collection, the blood collection volume for the next seven days will be reduced, and the part exceeding the predicted demand will be transferred to the ischemic and anemic areas. 5. The blood management decision support system according to claim 1, characterized in that: it also includes an emergency demand response module; The emergency demand response module includes a rapid response model for rapid identification, early warning and decision-making in emergency situations; Emergency demand response includes the following steps: S1. Collect and organize emergency situations that have occurred in history, including blood shortages, natural disasters and large-scale accidents, and analyze the characteristics of emergencies and response measures; S2. Use statistical and machine learning techniques to assess the probability of different emergency situations and establish a prediction model; S3. Develop an emergency list based on historical data and risk assessment; S4. For each emergency situation in the list, pre-set response measures; S5. Use real-time data monitoring and automated algorithms to quickly identify emergency situations; S6. When an emergency is detected to be about to happen or has already happened, the system can automatically issue an early warning; S7. Provide decision support tools to help managers make decisions quickly, including the selection and implementation of preset measures. 6. According to claim 2, a blood management decision support system is characterized in that: an LSTM model is used for prediction, which is expressed as: Where i = 1, 2, …, T; o, h and c represent the output, hidden state and cell state of LSTM respectively, T is the length of the time series window, and _ep represents the blood consumption after integrating all months in the time series window. 7. According to claim 3, a blood management decision support system is characterized in that: the blood collection prediction method includes the following steps: sorting historical blood collection data, checking the stationarity of the time series, and performing a differential operation when the time series is not stationary; performing an autocorrelation test on the data to determine how many periods of past data are used in the model to predict the current value; performing a partial autocorrelation test on the data to determine how many periods of residuals are used in the model to predict the current value; using ARIMA model parameters to fit the model to a portion of the data, and using the remaining data to evaluate the model; using the root mean square error to evaluate the degree of model fit, and using the trained ARIMA model to predict the blood collection data for the next seven days. 8. According to claim 1, a blood management decision support system is characterized in that: the blood use effect evaluation module takes the blood use volume as the research object and data set, selects the factors affecting the blood use volume prediction as the variable target, and establishes a logistics regression model; The blood use assessment includes the following steps: first, factor analysis is performed on the prediction indicators, the data is reduced in dimension, the correlation between the data is eliminated, and the final indicators are determined; then a logistics regression model is established based on the final indicators; finally, the prediction effect of the logistics regression model is tested, and suggestions for rational blood use are made based on the actual results. 9. According to claim 8, a blood management decision support system is characterized in that: the Logistics regression model is expressed as: Among them, _x1 is blood type A, _x2 is blood type B, _x3 is blood type AB, and _x4 is blood type O; suppose the dependent variable is a binary variable, with the value y=0 or y=1, and the m variables that affect the value of y are _{x1 x2 x3 x4} ; under the influence of m independent variables, the conditional probability of the positive result y=1 is p=p(y=1| _x1 , _x2 ,… _xm ); Perform logit transformation on the Logistics regression model expression to obtain the following linear form of the Logistics regression model: The constant term β is the natural logarithm of the ratio of the probability of adequate blood supply to that of inadequate blood supply when all exposure factors are 0; the partial regression coefficient β _j (j = 1, 2, 3, ..., m) represents the average change in logit (p) when the jth independent variable changes by one unit under the condition that other independent variables are fixed, and it has a corresponding relationship with the odds ratio OR; when other influencing factors are the same, the natural logarithm of the odds ratio of two different exposure levels of a risk factor x _j and c ₁ and c ₀ is: so: OR _j = exp[β _j (c ₁ −c ₀ )] The larger the OR, the greater the probability of a positive result, indicating that the blood supply for that blood type is sufficient.