CN114038563A

CN114038563A - Clinical machine withdrawal prediction system and method

Info

Publication number: CN114038563A
Application number: CN202111331807.XA
Authority: CN
Inventors: 杨旻; 刘万军; 刘瑜; 张金; 肖文艳; 许耀华; 华天凤; 李惠
Original assignee: Second Affiliated Hospital of Anhui Medical University
Current assignee: Second Affiliated Hospital of Anhui Medical University
Priority date: 2021-11-11
Filing date: 2021-11-11
Publication date: 2022-02-11

Abstract

The invention discloses a clinical machine withdrawal prediction system, which at least comprises: the clinical dynamic monitoring module is used for acquiring dynamic index data of clinical patients; the symptom index acquisition module is connected with the clinical dynamic monitoring module and acquires symptom performance index data according to clinical symptoms and patient data in a global medical database; the index data calculation module is connected with the symptom index acquisition module and constructs an optimal analysis model according to the symptom performance index data; and the display module is connected with the index data calculation module and the clinical dynamic monitoring module, and establishes a treatment dependence graph according to the optimal analysis model so as to guide withdrawal. The method can help clinicians to accurately predict the withdrawal of the machine for patients with sepsis.

Description

Clinical machine withdrawal prediction system and method

Technical Field

The invention belongs to the field of medical equipment control, and particularly relates to a clinical machine withdrawal prediction system and a clinical machine withdrawal prediction method.

Background

Sepsis is a syndrome of organ dysfunction resulting from a dysregulated body response to inflammation. It is a type of patient with rapid disease change and high mortality. Existing studies have demonstrated that patients have very high susceptibility of their lungs, and too long or too short mechanical ventilation is detrimental to the prognosis of sepsis patients. Therefore, in sepsis patients undergoing invasive ventilation, selecting the appropriate timing to withdraw the machine is extremely important for the prognosis of sepsis patients. The selection of the machine withdrawal parameters and the selection of the machine withdrawal time have great randomness in the current clinic, uniform guidance similar to guidelines does not exist, and corresponding machine withdrawal prediction analysis is not used as guidance opinions.

Disclosure of Invention

The invention aims to provide a clinical machine withdrawal prediction system and a clinical machine withdrawal prediction method, which are used for establishing a complete and feasible machine withdrawal scheme for patients with sepsis so as to help clinicians to accurately predict the machine withdrawal of the patients with sepsis.

In order to solve the technical problems, the invention is realized by the following technical scheme:

the invention provides a clinical withdrawal prediction system, which at least comprises:

the clinical dynamic monitoring module is used for acquiring clinical dynamic index data of the patient;

the symptom index acquisition module is connected with the clinical dynamic monitoring module and acquires symptom performance index data according to clinical symptoms and patient data in a global medical database;

the index data calculation module is connected with the symptom index acquisition module and constructs an optimal analysis model according to the symptom performance index data; and

and the display module is connected with the index data calculation module and the clinical dynamic monitoring module, and establishes a treatment dependence graph according to the optimal analysis model so as to guide withdrawal.

In an embodiment of the present invention, the clinical machine withdrawal prediction system further includes an online evaluation module, connected to the display module, the index data calculation module and the clinical dynamic monitoring module, where the online evaluation module calculates and obtains a simplified model and a machine withdrawal success rate according to the optimal analysis model to guide machine withdrawal.

In one embodiment of the invention, the symptom performance indicator data comprises biological indicator data and treatment data prior to weaning.

In one embodiment of the invention, the biological indicator data includes arterial blood gas, arterial partial pressure of carbon dioxide, excess base, whole blood count, hemoglobin, platelets, laboratory indicators, vital signs, temperature, and urine volume.

In one embodiment of the invention, the treatment data includes tidal volume, positive end expiratory pressure, number of invasive ventilation days, number of antibiotic use days, number of continuous renal replacement therapy days, and usage of vasopressors over 24 hours.

In one embodiment of the present invention, the index data calculation module includes:

the modeling unit is used for establishing a mathematical model according to the symptom performance index data;

the model analysis unit is used for drawing a test subject working curve chart of the mathematical model and acquiring working data of the test subject working curve chart; and

and the judging unit is used for obtaining the optimal analysis model according to the working data.

In one embodiment of the invention, the working data includes area under the curve, sensitivity, specificity, positive predictive value, negative predictive value, and model optimal cutoff value in the subject's working curve.

In one embodiment of the invention, the parametric information of the simplified model comprises mechanical ventilation time, positive end expiratory pressure value, urine volume, and residual alkalinity value.

In one embodiment of the invention, the parametric information for the therapy dependency map includes urine volume, lowest alkali excess, glasgow coma index, lowest blood oxygen saturation, congestive heart failure, lowest acid-base value, lowest mean arterial pressure, highest partial pressure of carbon dioxide, kidney disease, lowest platelets, body mass index, and oxygenation index.

A clinical setting-out prediction method is based on the clinical setting-out prediction system, and comprises the following steps:

obtaining symptom performance index data of the clinical symptoms in a global medical database according to the clinical symptoms of the patient;

constructing an optimal analysis model according to the symptom performance index data;

establishing a treatment dependence graph according to the optimal analysis model; and

and acquiring clinical dynamic index data of the patient, and guiding withdrawal according to the treatment dependence graph.

As mentioned above, the clinical withdrawal prediction system of the invention provides direct guidance of index parameter information for the clinician to guide withdrawal. The invention is provided with a symptom index acquisition module, acquires index data of patients who have done invasive mechanical ventilation sepsis all over the world, and updates the data in real time, so that the success rate judgment of the machine withdrawal of the invention is consistent with the current medical progress. The invention is provided with an index characteristic calculation module, and can select an optimal analysis model for calculating the relationship between the machine withdrawal success rate and the patient index parameter information through modeling combination. According to the invention, the treatment dependence graph is constructed in the form of a contribution scatter diagram by arranging the display module according to the optimal analysis model, so that a clinician can adjust and select an optimal withdrawal scheme. The invention is provided with an online evaluation module, calculates the machine withdrawing success rate through a simplified model and provides direct guidance for the withdrawing of the patient.

Of course, it is not necessary for any product in which the invention is practiced to achieve all of the above-described advantages at the same time.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.

Fig. 1 is a flowchart of a clinical setting-down method according to an embodiment of the present invention.

FIG. 2 is a schematic flow chart of obtaining a simplified model.

Fig. 3 is a schematic structural diagram of a symptom index acquisition module.

FIG. 4 is a working curve of a subject in an internal validation set with the modeling assembly of the present invention.

Figure 5 is a working curve of a subject in an external validation set with the modeling of the present invention.

FIG. 6 is a schematic diagram of an online evaluation module.

FIG. 7 is a SHAP merit analysis plot of index variables in the optimal analysis model.

FIG. 8 is a SHAP importance ranking graph of index variables in an optimal analysis model.

FIG. 9 is a calibration graph of an optimal analysis model and a simplified model.

FIG. 10 is a decision graph of an optimal analysis model and a simplified model.

Fig. 11 is a SHAP dependence of mechanical ventilation time.

FIG. 12 is a SHAP dependence of positive end expiratory pressure.

FIG. 13 is a SHAP dependence of urine output.

FIG. 14 is a SHAP dependence of the amount of base remaining.

FIG. 15 is a SHAP dependence of oxidation index.

FIG. 16 is a SHAP dependency graph of the Glasgow coma index.

FIG. 17 is a SHAP dependence graph of the length of time the antibiotic was used.

Fig. 18 is a SHAP dependency graph of the lowest blood oxygen saturation.

FIG. 19 is a SHAP dependence of peak body temperature.

Fig. 20 is a SHAP dependency graph of the highest heart rate.

FIG. 21 is a SHAP dependence graph of lowest pH.

FIG. 22 is a SHAP dependency graph of patient age.

FIG. 23 is a flowchart of a clinical withdrawal prediction method according to the present invention.

Table 1 is a table of predicted performance values in an internal validation set for the modeling assembly of the present invention.

Table 2 is a table of predicted performance values for the modeling assembly of the present invention in an external validation set.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Mechanical ventilation is a ventilation mode that utilizes mechanical means to replace, control, or modify spontaneous respiratory motion. Mechanical ventilation is an important treatment means in clinical treatment, for example, with the help of a ventilator, to maintain the airway of a patient unobstructed, improve ventilation and oxygenation, prevent hypoxia and carbon dioxide accumulation in the body, and create conditions for the body to possibly overcome respiratory failure caused by basic diseases and for treating the basic diseases. Although mechanical ventilation can save lives, a high risk of complications must be the early detachment of the patient from the ventilator, a process that withdraws the ventilator. In clinical treatment, ventilator withdrawal is an important clinical problem. For patients with sepsis, for example, mechanical ventilation can help them maintain proper ventilation, reduce respiratory muscle work, improve lung gas exchange function, and improve the pressure-volume relationship of the lungs. After the primary morbidity of a patient who is subjected to mechanical ventilation is controlled, ventilation and ventilation functions are improved, the support of the mechanical ventilation for breathing is gradually withdrawn, and the patient can recover complete spontaneous breathing, so that the successful withdrawal process is realized.

In most patients, mechanical ventilation can be stopped once the root cause of acute respiratory failure is resolved. However, there are also patients who have difficulty escaping from mechanical ventilation, for example 20% to 30%. When the patient fails the spontaneous breathing test within 48 hours after extubation or needs to be intubated again, the patient is judged to fail to withdraw the machine, and the patient suffers considerable risks due to breathing unsmooth after extubation and intubation again. Premature withdrawal may lead to failure of withdrawal, increasing reinsertion and mortality. Withdrawal is performed too late, which can lead to mechanical ventilation complications and can also result in excessive medical costs. Therefore, proper withdrawal regimens are also of paramount importance for the treatment of a patient's condition.

Referring to fig. 1, the present invention provides a clinical withdrawal prediction system 10, and the clinical withdrawal prediction system 10 includes a clinical dynamic monitoring module 101, a symptom index collecting module 102, an index data calculating module 103, a display module 104, and an online evaluation module 105. The clinical dynamic monitoring module 101 is connected to the clinician end 60, and is configured to obtain dynamic index data of the clinical patient before the preparation for withdrawal. The symptom index collection module 102 is connected to the clinical dynamics monitoring module 101, and is configured to collect symptom performance index data of the patient. The index data calculation module 103 is connected to the symptom index acquisition module 102, and is configured to obtain a mathematical model for evaluating the comprehensive status of the patient. And the display module 104 is connected to the index data calculation module 103 and the clinical dynamic monitoring module 101 and is used for constructing a visual analysis chart for guiding withdrawal. And the online evaluation module 105 is connected to the display module 104, the index data calculation module 103 and the clinical dynamic monitoring module 101, and is used for judging the machine withdrawal probability of the clinical patient.

Referring to fig. 1, the symptom indicator collecting module 102 collects symptom performance indicator data of clinical symptoms according to the clinical symptoms and patient data in the global medical database, and updates the symptom performance indicator data in real time. The index data calculation module 103 constructs an optimal analysis model according to the symptom performance index data. The display module 104 builds the treatment dependency graph 50 according to the optimal analysis model, so that a clinician adjusts the dynamic index data of the clinical patient according to the treatment dependency graph 50 to guide the clinical patient to withdraw from the machine. The online evaluation module 105 obtains a simplified model according to the optimal analysis model, and outputs a machine withdrawal success rate according to the simplified model and the dynamic index data of the clinical patient, so that a clinician can guide the machine withdrawal according to the machine withdrawal success rate.

Referring to fig. 2 and 3, in one embodiment of the present invention, the symptom index collection module 102 comprises a data collection unit 1021 and a screening unit 1022, wherein the data collection unit 1021 is used for collecting a global symptom index characteristic, such as a sepsis performance characteristic. The screening unit 1022 is used for screening collected data that do not meet the requirement of removing the machine. The data acquisition unit 1021 collects performance characteristics of patients with the same symptom in a global scope, so that the index data calculation module 103 performs modeling analysis on the collected performance characteristics to obtain a relationship between the patient's machine withdrawal success rate and related indexes, so that the online evaluation module 105 can directly calculate the machine withdrawal success rate of the patient through the simplified model, or construct the treatment dependency graph 50 by analyzing dynamic index data of clinical patients to guide the machine withdrawal behaviors of clinicians. The data collection source of The data collection Unit 1021 includes a first Database, such as an Intensive Care Medical Database MIMIC-IV, commonly known as Medical Information mark for Intensive Care Car-IV, and a second Database, such as an Emergency Intensive Care room Collaborative Research Database elcu-CRD, commonly known as The elcu Collaborative Research Database, wherein The elcu is commonly known as an Emergency Intensive Care Unit, i.e., Emergency Intensive Care room. Obtaining symptom manifestation characteristics of patients with the same symptom through the first database and the second database.

Referring to fig. 2 and 3, in one embodiment of the present invention, the data collected by the data collection unit 1021 also includes a release document published each year, and the established keywords include release, mechanical ventilation, ventilator, sepsis. Through the correlation analysis OF the keywords, documents related to the keywords are acquired by a crawler means in a scientific index website such as WEB OF SCIENCE, and the related documents are clustered and subjected to visualization analysis. The screening unit 1022 selects the relevant documents collected by the data acquisition unit 1021, the contents of the clinical common indicators related to the prognosis of the patient, and incorporates the selected contents into the symptom performance characteristics. The symptom index acquisition module 102 keeps updating the symptom performance index data, so that the clinical machine withdrawal prediction system 10 of the invention can keep up with the progress of disease research all the time, and has higher accuracy.

Referring to fig. 2 and 3, in one embodiment of the present invention, the symptom index characteristics collected by the data collection unit 1021 include the age, sex, and body mass index of the patient, and the biological index and treatment data before weaning, for example, 24 hours. The worst index data before the machine withdrawal, for example, within 24 hours, is selected as the biological index. The biological indicators include arterial blood gas, arterial carbon dioxide partial pressure (PaCO)₂) Base Excess (BE), whole blood count, laboratory indices, vital signs, temperature and urine volume. The index of arterial blood gas is represented by blood pH value (pH value) and arterial oxygen partial pressure (PaO)₂). The indices of the complete blood count are expressed as white blood cell count (WBC), Hemoglobin (HB), Platelets (PLT). Laboratory indices include creatinine, anion space. The indicators of vital signs are heart rate, respiratory rate, mean arterial pressure, peripheral blood oxygen saturation (SpO)₂). The treatment data comprises tidal volume, positive end expiratory pressure, invasive ventilation days, antibiotic usage days, blood vessel pressure increasing drug usage within 24 hours, and Continuous Renal Replacement Therapy (CRRT)element therapy) days.

Referring to fig. 1-3, in an embodiment of the invention, the data information collected by the data collecting unit 1021 is used to distinguish the symptom index feature according to individual patient. The screening method of the screening unit 1022 includes first determining whether the patient has experienced invasive mechanical ventilation, second determining whether the patient is under a certain age, such as 18, such as 20, 25, etc., and finally determining whether the patient has repeatedly entered the intensive care unit. In the raw data acquired by the data acquisition unit 1021 from the first and second databases, data of a patient who has undergone invasive mechanical ventilation, aged for example 18 years, first entered the intensive care unit is retained. The conditional sifting by the sifting unit 1022 obtains a first data set derived from the first database and a second data set derived from the second database. The screening unit 1022 divides the data in the first data set into the training set 20 and the internal verification set 30 according to a certain ratio, for example, 4:1, or 3:1, 5:1, etc. in units of individual patients. The data in the second data set is then all divided into an outer validation set 40. The screening unit 1022 passes the processed data to the metric data calculation module 103 in the form of data clusters of the training set 20, the internal validation set 30, and the external validation set 40.

Referring to fig. 1-3, in this example, sepsis patient data was collected, for example 10832 raw data in the first database and 33790 data in the second database. 7630 patients undergoing invasive mechanical ventilation in the first database and 7549 patients undergoing invasive mechanical ventilation in the second database, for example. The number of patients in the first data set is, for example, 5020 patients, and the number of patients in the second data set is, for example, 7081 patients, after the screening by the screening unit 1022. The first data set is divided in a ratio of, for example, 4:1, resulting in a training set 20 with a number of patients of, for example, 4016 cases and an internal validation set 30 with a number of patients of, for example, 1004 cases. The number of patients in the external verification set 40 is 7081. The index data calculation module 103 performs modeling according to the patient data in the training set 20, and the modeling method includes XGBoost, an artificial neural network, a random forest, a support vector machine, a K-value neighbor method, and a logistic regression method. The index data calculation module 103 performs model verification based on the data of the internal verification set 30 and the external verification set 40, and selects an analysis model with the highest accuracy. The behavior of each model in the internal verification set 30 is shown in table 1, and the behavior of each model in the external verification set 40 is shown in table 2.

TABLE 1

Referring to table 1 and fig. 1, in an embodiment of the present invention, the modeling methods used include an XGBoost model, an ANN model, an RF model, an SVM model, an LR model, and a KNN model, and these modeling methods form a modeling combination. In other embodiments of the invention, the approximate algorithm models of the models can be replaced to form a new modeling combination. The simplified model is a model obtained by simplifying The optimal analysis model in this embodiment by The online evaluation module 105, and will be described in detail later. In the modeling method of this embodiment, the XGBoost model, that is, the Extreme Gradient Boosting model, is collectively called Extreme Gradient Boosting. The ANN model is an Artificial Neural Network model and is called as an Artificial Neural Network. The RF model is a Random Forest model and is called Random Forest entirely. The SVM model is a support vector machine model, and is fully called support vector machines. The LR model is a Logistic Regression model, and is called Logistic Regression. The KNN model is a K value Neighbor method model and is totally called as K-Nearest Neighbor.

As shown in table 1, in the XGBoost model, each iteration is to add a tree to fit a residual between a predicted result and a true value of a previous tree based on an existing tree, so that the predicted value accuracy is high, and efficient learning of the XGBoost model also enables the XGBoost model to adapt to operation of a large amount of data. The ANN model can change information input and output by gradually adjusting the weight of the neurons on the basis of known data, so that the ANN model has obvious advantages in the aspects of processing fuzzy data, random data and nonlinear data, and is suitable for systems with large scale, complex structure and ambiguous information. The RF model selects the optimal characteristics on a common decision tree as the left and right sub-tree division of the decision tree, so that the RF model has good effect on the aspect of processing parallel data. The SVM model is a nonlinear classifier, and is favorable for solving an optimal solution of convex quadratic programming. The LR model has great advantages in solving and distinguishing the two classification problems. The KNN model is a parameter-free model, and can achieve good effect on the instant analysis of the data set because the model can achieve good effect in less training time. In this embodiment, the modeling combination is selected, so that the index calculation module 103 can fully consider the requirements of large-scale data size of medical data, parallelism of multiple medical index items, accuracy of model fitting, complex relationship network between related medical data, real-time analysis on continuously updated data, and the like during modeling work. The clinical withdrawal prediction system 10 background can handle the constant updating of medical data. In the modeling combination in this embodiment, the index feature calculation module 103 continuously calculates and selects the optimal analysis model along with the data update of the data acquisition unit 1021, so that the prediction system of the present invention always maintains high accuracy and fast data processing capability, and provides sufficient evidence for the clinician to withdraw the machine.

TABLE 2

Referring to fig. 2, 4 and 5, and tables 1 and 2, in one embodiment of the present invention, graphs of the work of the subjects for the different models in the modeled combination are plotted based on the data in the inner validation set 30 and the outer validation set 40. The abscissa of the test subject working curve graph is the false positive rate, the ordinate is the true positive rate, the area under the curve represents the sensitivity of the curve, and the larger the area under the curve is, the higher the judgment value of the curve is also indicated. The AUROC in tables 1 and 2 is collectively referred to as the Area Under the ROC currve, i.e., the Area Under the line of the subject's working Curve. In fig. 4, a false positive rate of, for example, 0.2 is taken as a boundary, and when the false positive rate is, for example, 0.2, the true positive rate values of the modeling composition according to the present invention are sorted from large to small into the XGBoost model, the simplified model, the LR model, the RF model, the ANN model (MLP model, Multilayer Perceptron, artificial neural network model), the SVM model, and the KNN model. In fig. 5, when the false positive rate is, for example, 0.2, the true positive rate values of the modeling combination of the present invention are sorted from large to small into the XGBoost model, the LR model, the simplified model, the ANN model, the RF model, the SVM model, and the KNN model.

Referring to fig. 2, fig. 4 and fig. 5, and tables 1 and 2, AUROC values, sensitivities, specificities, positive predictive values, negative predictive values, and model optimal cutoff values of model curves were obtained from the subject working curve charts of fig. 4 and fig. 5. In table 1, it can be seen that the model with the highest AUROC value is the XGBoost model, the area under the curve can reach, for example, 0.80, and the AUROC value ranges from, for example, 0.77 to 0.82. In table 2, it can be seen that the model with the highest AUROC value is the XGBoost model, which can reach, for example, 0.86, and the AUROC value ranges from, for example, 0.85 to 0.87. Meanwhile, the XGBoost model also performs well on sensitivity and specificity, positive predictive value and negative predictive value, and in table 1 and table 2, it can be seen that in the item of sensitivity, the SVM model can achieve a sensitivity of, for example, 0.59, in the range of, for example, 0.55 to 0.64. The SVM model can achieve a positive predicted value of 0.96, for example, and a positive predicted value range of 0.96-0.97, but the SVM model has a poor negative predicted value, so that the AUROC value is taken as a main consideration, and in the embodiment, the optimal analysis model can be the XGboost model. In other embodiments of the present invention, in the case of updating the first data set and the second data set, in the modeling combination of the present invention, the optimal analysis model may also be a model other than the XGBoost model. Through the modeling combination, the characteristics of different models can be applied, and the optimal analysis model can be flexibly obtained by combining the common verification of the internal verification set 30 and the external verification set 40.

Referring to fig. 1, after the index data calculation module 103 obtains the optimal analysis model through calculation, the optimal analysis model is transmitted to the online evaluation module 105. The online evaluation module 105 uses the symptom index features as variables according to the optimal analysis model and the symptom index features, and ranks and interprets the importance of the variables. The online evaluation module 105 simplifies the optimal analysis model to obtain a simplified model.

Referring to fig. 6, the online evaluation module 105 includes a simplification unit 1051, an information receiving unit 1052, and a calculation unit 1053. The simplifying unit 1051 is configured to receive the optimal analysis model, sort and screen model variables, select simplified decision variables, and obtain the simplified model according to the decision variables. The information receiving unit 1052 is connected to the clinical dynamics monitoring module 101, and receives the patient dynamics index data of the currently-prepared weaning machine acquired by the clinical dynamics monitoring module 101. The calculation unit 1053 picks out the decision variables in the dynamic index data of the patient, introduces the decision variables into the simplified model, and calculates to obtain the success rate of removing the computer.

Referring to fig. 6, since the optimal analysis model, such as the XGBoost model, is used to achieve high reliability determination of patient index, the data size involved is large and the index variable is large. Therefore, in this embodiment, the simplifying unit 1051 is provided to eliminate the low-related index variable that affects the success rate of the shutdown. The simplifying unit 1051 first uses the index variable of the optimal analysis model as an initial feature subset, selects n features, for example, to input the n features into a random forest classifier, calculates the importance of each feature, and obtains the classification accuracy of the initial feature subset of the device by using a cross validation method. The simplifying unit 1051 removes a feature with the lowest feature importance from the current feature subset to obtain a new feature subset, inputs the new feature subset into the random forest classifier again, calculates the importance of each feature in the new feature subset, obtains the classification accuracy of the new feature subset by using a cross validation method, and repeats the steps. Until the current feature subset is empty, a total of feature subsets with different feature quantities, such as k, are obtained, and the feature subset with the classification precision ordered in the top 4 bits, for example, is selected as the optimal feature combination.

Referring to fig. 7 and 8, the abscissa in fig. 7 is the SHAP (SHapley Additive exposition) value of the symptom characteristic index output by the model, and represents the importance of the index variable. The abscissa in fig. 8 ranks the importance of each index variable in the optimal analysis model. The optimal feature combination has the highest contribution to the optimal analysis model, and in this embodiment, the optimal feature combination includes mechanical ventilation time, positive end expiratory pressure value, urine volume, and base remaining value. In fig. 7, the probability of machine withdrawal success increases as index data including urine volume, minimum alkali excess, glasgow coma index (GCS), minimum oxygen saturation of blood (SPO) increases₂) Congestive heart failure, minimum acid-base number (pH), minimum mean arterial pressure, maximum partial carbon dioxide Pressure (PCO)₂) Kidney disease, lowest platelet, Body Mass Index (BMI), and oxygenation Index.

Referring to fig. 2, fig. 6, table 1 and table 2, based on the optimal feature combination, such as mechanical ventilation time, positive end expiratory pressure, urine volume and residual alkalinity, a simplified unit 1051 calculates the simplified model, and then compares the simplified model with the models in the modeling combination based on the data in the internal validation set 30 and the external validation set 40 to obtain the AUROC value, sensitivity, specificity, positive predictive value, negative predictive value and the optimal model cutoff value of the simplified model. And drawing a calibration curve and a decision curve of the simplified model and the optimal analysis model.

Referring to fig. 9 and 10, the abscissa in fig. 9 is the predicted risk of the model, and the ordinate is the observation frequency of the judgment sample of the model. The abscissa in fig. 10 is the high risk threshold and the cost benefit ratio, and the ordinate is the net benefit. The simplified model has a boolean score of, for example, 18.5, and a boolean score in the range of, for example, 18.0 to 19.1. The Boolean score of the optimal analysis model is 9.5 for example, the value range of the Boolean score is 8.9-10.1 for example, the simplified model is calibrated through the Boolean score, the stability of the simplified model is good, and the machine withdrawal success rate calculated through the simplified model is credible. The net gain of the simplified model and the optimal analytical model is higher for the optimal analytical model at each high risk threshold probability, while the difference in net gains between the simplified model and the optimal analytical model is smaller for thresholds less than, for example, 0.4, and the simplified model also has higher gains.

Referring to fig. 8, the display module 104 receives the optimal analysis model from the index feature calculation module 103, and according to the importance of each index variable in fig. 8, and according to the continuous index variable therein, draws a treatment dependency graph 50, where the treatment dependency graph 50 may be a graph for judging the contribution degree of the continuous index variable to the withdrawal success rate. In this embodiment, the treatment dependency graph 50 may be a SHAP scatter plot to account for the effect of changes in the value of each variable in the optimal analytical model on weaning patients. Without reference to the simplified model, the present invention provides a treatment dependency graph 50 to facilitate the clinician in making a customized evacuation plan based on each variable.

Referring to fig. 1, 11-22, in the treatment dependency graph 50 established by the display module 104, the abscissa is the index variable, and the ordinate is the contribution to the machine withdrawal success rate. Wherein the blue scatter points represent characteristic values and the red curves represent linear relations of horizontal and vertical coordinates. The index variables involved in the contribution include mechanical ventilation time, positive end expiratory pressure, urine output, residual alkalinity, highest negative ion space, glasgow coma index, duration of antibiotic use, lowest blood oxygen saturation, highest body temperature, highest heart rate, lowest ph, and patient age. Wherein, the index variables which are positively correlated with the machine withdrawal success rate comprise urine output, alkali residual quantity, Glasgow coma index, lowest blood oxygen saturation and lowest pH value. Index variables negatively correlated with the machine withdrawal success rate include mechanical ventilation time, positive end expiratory pressure, maximum negative ion clearance, antibiotic usage duration, age. Index variables that are positively or negatively correlated with the machine withdrawal success rate, including the highest body temperature and the highest heart rate, cannot be judged significantly.

Referring to fig. 1 and 11, in an embodiment of the present invention, when the mechanical ventilation time is, for example, 0.1 to 1.5 days, the machine removal success rate is high, and particularly when the mechanical ventilation time is, for example, 1 day, the machine removal success rate may reach a peak, and the SHAP contribution degree thereof may reach, for example, 0.4. Mechanical ventilation time has a negative effect on the withdrawal success rate after mechanical ventilation time of more than, for example, 2 days, and mechanical ventilation time has a smaller effect on the withdrawal success rate after mechanical ventilation time of more than, for example, 5 days, and the SHAP contribution degree is maintained at, for example, -0.3. Therefore, when the clinician guides the machine withdrawal according to the treatment dependence graph 50 of the display module 104, the mechanical ventilation time of the patient is preferably selected to be within 0.1-1.5 days, and when the mechanical ventilation time of the patient is more than 5 days, for example, other index variable graphs are introduced as the machine withdrawal guide. Alternatively, the clinician may control the patient's withdrawal time to within, for example, 2 days of mechanical ventilation.

Referring to FIGS. 1 and 12, in one embodiment of the present invention, the positive pressure is 0.1-5 cmH at the end of expiration₂Within O, the success rate of machine withdrawal is high, especially when the positive pressure at the end of expiration is, for example, 4cmH₂O, the machine withdrawal success rate can reach a peak, and the SHAP contribution can reach, for example, 0.4. When the expiratory end pressure is above e.g. 8cmH₂After O, the end-expiratory pressure has a negative effect on the success rate of withdrawal when it is greater than, for example, 15cmH₂post-O, positive end expiratory pressure has less effect on weaning success, with the SHAP contribution remaining at, for example, -0.4. Therefore, when the clinician guides the machine withdrawal according to the treatment dependency graph 50 of the display module 104, the positive pressure at the end of the expiration of the patient is preferably 0.1-5 cmH₂In O, when the patient's terminal expiratory pressure is greater than, e.g., 15cmH₂And when O is needed, introducing other index variable graphs as the machine withdrawal guide. Alternatively, the clinician may control the patient's evacuation time to the end of expirationPositive pressure, e.g. less than 4cmH₂And O is in the range.

Referring to fig. 1 and 13, in one embodiment of the present invention, when the patient's urine output is less than, for example, 0.6ml/(kg × h), the success rate of weaning is negatively affected. When the urine output of a patient is greater than, for example, 1.8ml/(kg × h), the patient has a high weaning success rate and the SHAP contribution can reach, for example, 0.2. Therefore, when the clinician instructs withdrawal according to the treatment dependency graph 50 of the display module 104, the clinician preferentially confirms that the urine output of the patient is greater than, for example, 1.8ml/(kg × h), and adjusts the treatment plan when the urine output is less than, for example, 0.6ml/(kg × h), to temporarily stop withdrawal.

Referring to fig. 1 and 15, in an embodiment of the present invention, when the highest negative ion gap is within 1-12.5 mEq/L, the machine removal success rate is high, and the SHAP contribution degree can reach, for example, 0.2. The highest anion gap has a negative effect on machine withdrawal success rates when the highest anion gap is greater than, for example, 18 mEq/L. The machine withdrawal success rate continues to decrease as the maximum negative ion gap increases thereafter. Therefore, the clinician gives priority to the patient's highest anion gap of, for example, 1-12.5 mEq/L when instructing withdrawal of the machine based on the treatment dependency graph 50 of the display module 104. When the highest negative ion gap of a patient is larger than 18mEq/L for example, other index variable maps are introduced as withdrawal guidance. When the highest negative ion gap of the patient is larger than 20mEq/L, the treatment scheme is adjusted, and the machine is temporarily not withdrawn.

Referring to fig. 1 and 17, in one embodiment of the present invention, when the antibiotic is used for a period of time of, for example, 2 days, the disarming success rate is high, and the SHAP contribution can reach, for example, 0.2. When antibiotics are used for longer than, for example, 4.5 days, the length of antibiotic use has a negative effect on the success rate of weaning. The success rate of withdrawal continues to decrease as the length of antibiotic use increases thereafter. Therefore, the clinician gives priority to the patient's antibiotic usage for, e.g., 1-2 days when instructing withdrawal according to the treatment dependency graph 50 of the display module 104. When the patient has antibiotic usage for longer than, for example, 4.5 days, other index variograms are introduced as withdrawal guidelines.

Referring to fig. 1 and 19, in one embodiment of the present invention, when the patient has a maximum body temperature of 36.5-38 ℃, the patient has a high weaning success rate, and the SHAP contribution can reach, for example, 0.1. Therefore, when the clinician instructs withdrawal according to the treatment dependency graph 50 of the display module 104, when the body temperature of the patient is lower than 36.5 ℃ or higher than 38 ℃, other index variables are introduced to instruct withdrawal, or the treatment scheme is adjusted, and withdrawal is not performed for the moment.

Referring to fig. 1 and 20, in an embodiment of the present invention, when the patient's highest heart rate is, for example, 70-110 times/min, there is an improvement effect on the patient's success rate, wherein when the patient's highest heart rate is, for example, 95, the patient's machine withdrawal success rate is the highest, and the SHAP contribution degree can reach, for example, 0.15. Therefore, when the clinician instructs withdrawal according to the treatment dependency graph 50 of the display module 104, when the highest heart rate of the patient is less than 70 times/min or more than 110 times/min, other index variables are introduced to instruct withdrawal, or the treatment scheme is adjusted, and withdrawal is suspended.

Referring to fig. 2 and 23, the present invention provides a clinical setting-out prediction method based on the clinical setting-out prediction system 10, which includes the following steps.

S1, the module 102 for collecting indexes based on symptom obtains the data of the indexes of symptom expression of clinical symptoms in the global medical database according to the clinical symptoms of patients.

S2, an optimal analysis model is constructed according to the symptom performance index data based on the index data calculation module 103.

And S3, establishing a treatment dependence graph 50 according to the optimal analysis model based on the display module 104.

S4, obtaining a simplified model of the optimal analysis model based on the online evaluation module 105.

S5, based on the clinical dynamic index monitoring module 101, obtaining the clinical dynamic index data of the patient, calculating the machine withdrawal success rate of the patient by using the simplified model, and guiding the machine withdrawal according to the treatment dependency graph 50 and the machine withdrawal success rate.

In the description of the present specification, reference to the description of the terms "present embodiment," "example," "specific example," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

The embodiments of the invention disclosed above are intended merely to aid in the explanation of the invention. The examples are not intended to be exhaustive or to limit the invention to the precise forms disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, to thereby enable others skilled in the art to best utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A clinical withdrawal prediction system, comprising:

2. The clinical machine withdrawal prediction system according to claim 1, further comprising an online evaluation module connected to the display module, the index data calculation module and the clinical dynamic monitoring module, wherein the online evaluation module calculates and obtains a simplified model and a machine withdrawal success rate according to the optimal analysis model to guide machine withdrawal.

3. A clinical component withdrawal prediction system as claimed in claim 1 wherein the symptom performance indicator data comprises pre-component withdrawal biological indicator data and treatment data.

4. A clinical component withdrawal prediction system as claimed in claim 3 wherein the biological indicator data includes arterial blood gas, arterial partial pressure of carbon dioxide, excess base, whole blood count, hemoglobin, platelets, laboratory indicators, vital signs, temperature and urine volume.

5. The clinical component withdrawal prediction system of claim 3, wherein the treatment data comprises tidal volume, positive end expiratory pressure, number of invasive ventilation days, number of antibiotic use days, number of continuous renal replacement therapy days, and usage of vasopressors over 24 hours.

6. The clinical component withdrawal prediction system of claim 1, wherein the metric data calculation module comprises:

7. The clinical withdrawal prediction system of claim 6, wherein the working data comprises an area under the curve, sensitivity, specificity, positive predictive value, negative predictive value, and model optimal cutoff value in a working curve plot of the subject.

8. A clinical component withdrawal prediction system as claimed in claim 2 wherein the parametric information of the simplified model includes mechanical ventilation time, positive end-expiratory pressure, urine volume, and base residual value.

9. A clinical withdrawal prediction system according to claim 1, wherein the parametric information of the therapy dependency graph includes urine volume, minimum excess base, glasgow coma index, minimum blood oxygen saturation, congestive heart failure, minimum ph, minimum mean arterial pressure, maximum partial carbon dioxide pressure, kidney disease, minimum platelets, body mass index, and oxygenation index.

10. A clinical withdrawal prediction method, based on the clinical withdrawal prediction system of claim 1, comprising the following steps: