TWM648156U - System for predicting extubation using machine learning models - Google Patents

Abstract

This invention provides a system for establishing extubation prediction using a machine learning model, which can obtain an extubation prediction model and the key features used through the training and/or verification of a machine learning model, and through the extubation prediction The model analyzes a patient's key characteristic data in real time to obtain the patient's extubation possibility and related explanations. Accordingly, the system disclosed in this work using a machine learning model to establish extubation prediction is used as a tool for clinical caregivers to evaluate extubation, so as to reduce the possibility of reintubation due to inability to breathe spontaneously after extubation.

Description

Establishing a system for predicting extubation using machine learning models

This work relates to an extubation prediction system, specifically a system that uses a machine learning model to establish extubation prediction.

According to statistics, more than 80% of patients in intensive care units require mechanical ventilation to maintain their lives. Research shows that the longer a patient relies on mechanical ventilation, the harder it is to wean the patient off the mechanical ventilation. The difficulty of mechanical breathing assistance will increase, and it may even evolve into long-term dependence on mechanical breathing assistance to survive. Or, after being separated from the respirator for a short period of time, the patient must quickly be reintubated and rely on mechanical breathing assistance to survive.

In order to increase the success rate of patients weaning off mechanical respiratory assistance, many studies are currently using various research methods to predict the time point and success rate of patients weaning off mechanical respiratory assistance. For example, some studies are through reorganizing intensive care units and chronic respiratory care Statistical analysis was conducted on the patients in the ward and their clinical characteristics, reasons for using respirators, chronic complications, reasons for difficulty in weaning off respirators, etc., in order to summarize the characteristics of patients who are difficult to wean off mechanical respiratory assistance. However, the conditions of each patient are not the same. Moreover, the disease condition changes rapidly, and the results obtained through retrospective statistical analysis methods will not be able to accurately predict the time point suitable for extubation.

The main purpose of this creation is to provide a system for establishing extubation prediction using a machine learning model, which can automatically predict the possibility of extubation of inpatients in real time and serve as a tool for clinicians to evaluate the success rate of patient extubation.

Another purpose of this creation is to provide a system for establishing extubation prediction using a machine learning model, which provides reasons or related explanations for predicting the possibility of extubation, so as to improve the credibility of the possibility of extubation and Interpretability.

Therefore, in order to achieve the above purpose, this invention provides a system for establishing extubation prediction using a machine learning model, which includes a processing device for analyzing key characteristic data of a patient to be tested through an extubation prediction model. To generate the possibility of extubation of the patient to be tested, where the key characteristic data includes the physiological parameter data and the awareness obtained one day or/and two days before the date of extubation prediction. data, the input/output fluid data, the respiratory function data.

Among them, the extubation prediction model is selected from Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), Light Gradient Boosting Machine (LightGBM), Random Forest Algorithm ( Random Forest (RF) and logistic regression (Logistic Regression, LR).

Among them, the possibility of extubation is the possibility that the patient to be tested will be separated from the respiratory assistance device within 24 hours after the predicted date of extubation.

Among them, the processing device has a data processing module and an extubation prediction module, and the data processing module is used to extract the key characteristic data from the characteristic data of the patient to be tested; the extubation prediction module The group uses the extubation prediction model to analyze the key characteristic data.

In one embodiment of this invention, a system for establishing extubation prediction using a machine learning model further includes a database for collecting characteristic data of a plurality of patients, where the characteristic data includes a consciousness data, an input /Output liquid data, respiratory function data and physiological parameter data; The processing device includes a data processing module; and each patient has a record of using a respiratory assistance device during hospitalization.

In another embodiment of the present invention, the processing device is connected to the database, and the data processing module can be used to perform a data preprocessing procedure on the characteristic data of the patients in the database to Delete the data in the characteristic data that exceeds a preset reasonable range, and/or fill in the missing data in the characteristic data.

In another embodiment of the invention, the extubation prediction module uses the extubation prediction model to analyze the key characteristic data and generates correlation information between the key characteristic data and the extubation probability.

In a second embodiment of this invention, the processing device further includes a visualization module, which converts the correlation information between the key characteristic data and the probability of extubation into a visual interface, such as a bar chart or a broken line. graph, SHAP force value graph, partial dependency graph (PDP graph), etc.

In another embodiment of the present invention, the processing device further includes a model training module, which trains a machine learning model using at least part of the characteristic data of the patients, and can perform verification to generate the machine learning model. Extubation prediction model and one key feature.

Among them, the key features include age, number of days using a respirator, as well as GCS score, urine output, injection volume, nutrition amount, RASS score, PIP, MAP, respiratory rate 1 day and 2 days before the predicted date of extubation. Frequency, heart rate.

The actual implementation method of this creation to build a system for extubation prediction using a machine learning model includes the following steps: (a) Obtaining characteristic data for training; (b) Training with a machine learning model and obtaining an extubation training model; (c) Obtain key characteristic data of the patient to be tested; (d) analyze the key characteristic data using an extubation training model to obtain the extubation possibility of the patient to be tested.

Among them, step a further includes a data preprocessing step to remove those that do not meet a standard from the training feature data, and/or make up for the insufficient amount of the training feature data by interpolation.

Among them, step d further includes obtaining the correlation between each key feature and the extubation probability, and converting it into a visual interface.

10: Establish an extubation prediction system using machine learning models

20:Database

30: Processing device

31:Data processing module

32: Model training module

33: Extubation prediction module

34:Visualization module

Figure 1 is a schematic diagram of a system for establishing extubation prediction using a machine learning model according to an embodiment of the present invention.

Figure 2A is a flow chart (1) of the data analysis disclosed in this creation.

Figure 2B is a flow chart (2) of the data analysis disclosed in the present invention.

Figure 3A shows the performance of different machine learning models in predicting extubation using calibration curve analysis.

Figure 3B shows the performance of different machine learning models in predicting extubation using decision curve analysis.

Figure 3C shows the performance of different machine learning models in predicting extubation using the area under the curve.

Figure 4 shows the importance of each major clinical area to critical care.

Figure 5 illustrates the correlation between each key feature and the extubation prediction model using SHAP values.

Figure 6A is a partial dependence diagram of the influence of GCS on the predicted extubation probability by XGBoost.

Figure 6B is a partial dependence diagram of RASS affecting XGBoost’s prediction of extubation probability.

Figure 6C is a partial dependence diagram of the effect of urine volume on the predicted extubation probability by XGBoost.

Figure 6D is a partial dependence diagram of the injection volume affecting the predicted extubation probability by XGBoost.

Figure 6E is a partial dependence diagram of PIP affecting XGBoost’s prediction of extubation probability.

Figure 6F is a partial dependence diagram of MAP affecting XGBoost’s prediction of extubation probability.

Figure 7A is the result of Case 1 showing the overall impact of key features on the prediction of extubation for two different individuals through the LIME and SHAP force values of key features.

Figure 7B is the result of Case 2 showing the overall impact of key features on the prediction of extubation for two different individuals through the LIME and SHAP force values of key features.

This invention discloses a system that uses a machine learning model to establish extubation prediction. It can obtain an extubation prediction model and its key features through the training and/or verification of a machine learning model, and through the extubation prediction The model analyzes a patient's key characteristic data in real time to obtain the patient's extubation possibility and related explanations. Accordingly, the system disclosed in this work using a machine learning model to predict extubation is used as a tool for clinical caregivers to evaluate extubation, so as to reduce the possibility of being unable to breathe spontaneously or needing to be reintubated after extubation.

It must be emphasized that the system for establishing extubation prediction using machine learning models disclosed in this work incorporates patient consciousness data and input/output fluid data, so that the obtained extubation prediction results can be closer to what the doctor expects. The result of clinical judgment means that the system and method for establishing extubation prediction based on machine learning model disclosed in this work can achieve a more accurate extubation prediction rate.

The terminology used in this work will be explained below. If it is not listed in the description below, it will be explained based on the reference materials recognized by those with ordinary knowledge in the technical field of this work, such as dictionaries, dictionaries, literature or common knowledge.

The term "dynamic parameters" refers to the data or results obtained from regular or irregular testing of patients during their hospitalization.

The term "consciousness/cognition block" refers to data obtained based on systems, scales or methods for assessing a patient's state of consciousness, such as the Glasgow Coma Scale (GCS), RASS Sedation Assessment Scale (Richmond Agitation Sedation) Scale, RASS), etc.

The term "fluid balance block" refers to data related to the fluid entering the patient's body or the fluid eliminated by the patient, such as urine output, injection volume, infusion volume, and total nutrition.

The term "respiratory function block," refers to data related to the patient's respiratory function or cardiopulmonary function, such as peak airway pressure (PIP), mean airway pressure (MAP), number of ventilator days, and respiratory rate.

The term "physiological parameter block" refers to the patient's physiological data, such as weight, body mass index, height, etc.

The term "respiratory assistance equipment" or "respirator" refers to an external device installed on a patient to assist the patient in ventilating/inhaling.

The term "module" refers to a component with specific functions composed of several basic functional components, which can be used to form a complete functional system, device or program. For example, a module can be an electronic circuit.

The term "ICU", the full name is Intensive Care Unit, is an intensive care unit in a hospital, which is affiliated to different departments according to departments.

The term "machine learning" refers to using a machine learning model to learn and improve in data, find patterns and relationships, and formulate decisions and predictions based on its learning and analysis results. Taking the examples listed in this creation as an example, the machine learning model includes XGBoost (Extreme Gradient Boosting), CatBoost (Categorical Boosting), LightGBM (Light Gradient Boosting Machine), Random Forest algorithm (Random Forest, RF) and Logistic Regression (LR).

The "tube" in the term "extubation" refers to the tube used for ventilation or gas supply. Therefore, "extubation" means disconnecting from the respirator or other respiratory assistance equipment.

The term "APACHE II", full name acute physiology and chronic health evaluation II, is a disease severity scoring system, ranging from 0-71 points. The scoring system is based on 12 physiological data, patient age and health status. When entering the score, each data is the worst value of the patient 24 hours after admission.

The term "SOFA", the full name is Sequential Organ Failure Assessment, is a system for assessing the severity of a disease. It divides organs into lungs, coagulation, liver, heart, nerves, and kidneys, and evaluates them separately. The highest score for each organ is 4 points, with the lowest score being 0 points. The total score will be calculated every 48 hours after 24 hours after entering the ICU.

The term "GCS (Glasgow Coma Scale)" in Chinese is the Glasgow Coma Scale. It is a coma index and is one of the most widely used coma indexes. Its evaluation includes three aspects: eye opening reaction, speaking reaction, and movement. For each aspect, the maximum score is 5 points and the minimum score is 1 point. When the total score is lower, it means the coma procedure is more serious.

The term "RASS Sedation Scale (Richmond Agitation-Sedation Scale)" is a scale used to measure a patient's agitation or sedation level. The scores range from -4 to +4, representing different behavioral states.

The term "SHAP value", "SHAP force plot" or "SHAP value" means that when a predetermined machine learning model generates a predicted value for a data or set of data, the data or set of data The value assigned to each feature in .

The term "LIME (Local Interpretable Model-Agnostic Explanations)" refers to a locally interpretable model. Its purpose is to understand complex and difficult-to-interpret models, and the principle used is to provide a locally interpretable or interpretable model based on the individual to be explained. Models of understanding.

The term "visual interface" refers to the use of graphics to represent data or information in a graphical manner so that users can better understand the data and information. The so-called graphics include diagrams, tables, colors, symbols, Symbols, icons or other visual elements.

Specifically, please refer to Figure 1. One embodiment of the present invention provides a system 10 for establishing extubation prediction using a machine learning model, which includes a database 20 and a processing device 30, wherein: the database 20 is It is a structured information or data collection that will be stored in a predetermined way, such as electronically stored in a computer system, hard drive or cloud space. The database 20 collects and stores Store characteristic data of a plurality of patients, wherein each patient has a record of using a respiratory assist device during hospitalization in the intensive care unit, and the characteristic data includes a consciousness data, an input/output fluid data, and a respiratory function data. and a physiological parameter data. The data source of the database 20 can be directly input or obtained after being connected with other external databases.

In this embodiment, the consciousness data is the result of assessing the patient's consciousness or cognitive level, such as the score obtained by the Glasgow Coma Scale and the score obtained by the RASS Sedation Assessment Scale; the input/output liquid data is a predetermined time The amount of fluid injected into the patient's body, or the amount of fluid excreted by the patient, such as infusion volume, injection volume, liquid intake in the diet, urine output, etc.; the respiratory function data is related to the patient's respiratory or/and cardiopulmonary function status Relevant data, such as the number of days of wearing respiratory assistance equipment, peak airway pressure (PIP), mean airway pressure (MAP), respiratory frequency; the physiological parameter data is the patient's physiological data, such as age, weight, body mass index, etc.

The processing device 30 can be used as a series of hardware to execute programs, instructions or operate on data, such as computers, computer processors, etc. The processing device 30 is connected to the database 20 and has a data processing module 31, a model training module 32, an extubation prediction module 33 and a visualization module 34; wherein: the data processing module 31 It is used to process characteristic data from patients, including performing a data preprocessing process and a characteristic data retrieval process. Specifically, the data preprocessing procedure refers to the steps in which the data processing module 31 reads the characteristic data of the patients in the database 20 and then deletes or/and supplements the characteristic data. That is, the data processing module receives a preset reasonable range of each feature and uses it as a judgment standard. When a feature data does not fall within the preset reasonable range, the feature data is deleted; and the feature is checked. Whether the amount of data meets a predetermined value. If it does not meet the predetermined value, it is determined that the characteristic data is missing, and the missing part of the data should be filled in by taking the average of the characteristic data. The process of retrieving key features means that the data processing module 31 reads the characteristic data of a patient to be tested and extracts data related to a key feature as a key feature data.

The model training module 32 receives the processed feature data from the data processing module 31, trains a machine learning model with at least a part of the feature data, and verifies it with at least a part of the feature data. The trained machine learning model and its obtained results are used to produce an extubation prediction model and the key features used in the extubation prediction training model.

In this embodiment, the machine learning model is a well-known calculation logic or algorithm in the technical field of this creation, such as XGBoost (Extreme Gradient Boosting), CatBoost (Categorical Boosting), LightGBM (Light Gradient Boosting Machine), Random Forest Algorithm (Random Forest, RF) and logistic regression (Logistic Regression, LR), etc.

In this embodiment, there are 20 key features, including age, number of days of using respiratory assistance equipment, GCS scores, urine and urine measured one day and two days before the day of extubation prediction. Volume, injection volume, nutritional amount, RASS score, PIP, MAP, respiratory rate, heart rate.

The extubation prediction module 33 obtains the extubation prediction model from the model training module 32, and uses the extubation prediction model to analyze the key characteristic data from the patient to be tested to generate the patient to be tested in the The probability of extubation within 24 hours after the date of extubation prediction, and the correlation information between each key feature and the probability of extubation.

The visualization module 34 converts the correlation information between each key feature and the extubation probability into a visual interface for presentation on a display or a device with a display, such as a computer, screen, tablet, or mobile phone. wait.

Through the composition of the above components, this invention discloses a system that uses a machine learning model to establish extubation prediction, which can automatically analyze in real time whether patients hospitalized in the intensive care unit are in a state suitable for weaning from respiratory assistance equipment. In addition to using The quantitative method provides clinicians with the possibility of extubation as a reference for evaluating extubation, and can further provide clinicians with information on the correlation between each key feature and the possibility of extubation, thereby increasing the possibility of extubation. Explanatory, achieving the effect of increasing the credibility of the possibility of extubation.

Furthermore, the actual execution steps of the system that uses machine learning models to establish extubation prediction are as follows:

Step 101: Input a plurality of characteristic data for training, wherein the characteristic data comes from a plurality of patients who used respiratory assistance equipment during hospitalization in the intensive care unit, and the characteristic data includes a consciousness data, an input/output liquid data, A respiratory function data and a physiological parameter data, and the description of the consciousness data, the input/output liquid data, the respiratory function data and the physiological parameter data are the same as those described in the previous embodiment, so they will not be described again here. .

Step 102: Perform a data preprocessing procedure on the training feature data to delete unreasonable feature data and/or fill in the missing feature data; to complete the data preprocessing procedure of the training feature data. At least one part trains a machine learning model, and at least one part of the training feature data is able to complete the data preprocessing process and performs a verification process on the machine learning model that has completed the training to produce an extubation prediction model and an extubation prediction model. Key Features.

Among them, the extubation prediction model is selected from XGBoost (Extreme Gradient Boosting), CatBoost (Categorical Boosting), LightGBM (Light Gradient Boosting Machine), Random Forest algorithm (Random Forest, RF) and Logistic Regression (LR) The group formed.

Among them, the key features include age, number of days using a respirator, as well as GCS score, urine output, injection volume, nutrition amount, RASS score, PIP, MAP, respiratory rate 1 day and 2 days before the predicted date of extubation. Frequency, heart rate.

Step 103: Enter key characteristic data of a patient to be tested, wherein the patient to be tested is in a state of using a respiratory assistance device, and the key characteristic data is the key characteristic measured from the patient to be tested. data or numerical value.

Step 104: Use the extubation prediction model to analyze the key characteristic data, and generate the extubation probability rate of the patient to be tested within 24 hours after the extubation prediction date, as well as the relationship between the key characteristic data and the extubation possibility. The correlation information between the rates is converted into a visual interface, including colors, patterns, graphics, shapes, lines or a combination of at least any two of the above.

In the following, in order to illustrate the effects that the technical features of this invention can achieve, several examples will be given for detailed explanation.

All trials in the following examples were reviewed and approved by the Research Ethics Review Board of Taichung Veterans General Hospital, and followed the guidelines of informed consent and anonymity.

Example 1: Subject data analysis

Collect the data of patients admitted to the Intensive Care Unit of Taichung Veterans General Hospital from July 2015 to July 2019, and exclude the data of patients who are not using respirators, and the data of patients who have used respirators for less than 72 hours, and finally filter out the data of patients who have used respirators for less than 72 hours. A total of 5940 subjects were exposed to respirators for more than 48 hours. The statistical analysis results of the subject's characteristic data are shown in Table 1. The data in Table 1 are expressed as mean ± standard deviation or number (percentage). Presented in a way; CCI stands for Charlson Comorbidity Index. It can be seen from the contents of Table 1 that there were 3657 subjects who were weaned from the respirator during the ICU period, and 2283 subjects who were not weaned from the respirator during the ICU period; a total of 65 characteristic data were used; the average of the subjects The age was 66.2±16.2 years old, and 64.0% were male; the disease severity of the subjects was significantly higher, the APACHE II score and SOFA score were 25.7±6.6 and 8.5±3.6 respectively, and 61.5% of the subjects The patients were extubated during ICU hospitalization. However, the distribution of age, gender, and CCI index between extubated and non-extubated subjects was similar. However, the APACHE II scores and SOFA scores of non-extubated subjects were higher than those of extubated subjects.

Then, according to the four main clinical domains in the clinical workflow, the dynamic parameters of the subjects during their hospitalization in the ICU were classified and analyzed statistically. The results are shown in Table 2. Among them, the four The main clinical areas are consciousness/awareness domain, fluid balance domain, ventilatory function domain, and physiological parameter domain, among which , the consciousness/cognition block includes the Glasgow coma scale (GCS) and RASS sedation assessment. Richmond Agitation Sedation Scale (hereinafter referred to as RASS); the body fluid balance block includes the fluid, urine output, and total nutrition given to the patient; the respiratory function block includes peak airway pressure (PIP) and mean airway pressure (MAP) , ventilator days, respiratory rate; the physiological parameter block includes heart rate. From the results in Table 2, it can be seen that the consciousness of extubated patients during ICU hospitalization continued to improve, the sedation state decreased, the heart rate and infusion volume gradually decreased, and the urine The amount and nutrient content showed a steady increase.

Example 2: Results of machine learning model calculation

Please refer to Figure 2A. From the subject characteristic data collected in Example 1, the 20 most relevant features (hereinafter referred to as the 20 key features) are taken as modeling data and analyzed with different machine learning models. , among which, the machine learning models used include XGBoost (Extreme Gradient Boosting), CatBoost (Categorical Boosting), LightGBM (Light Gradient Boosting) Machine), Random Forest (RF for short) and Logistic Regression (LR for short), and the training/testing ratio is 80/20. The 20 key features are the result of analyzing the original feature data through recursive feature elimination analysis.

Please refer to Figure 2B. In order to achieve the purpose of prediction one day before extubation, classification is based on the time point when the subject characteristic data is obtained, which means that the data one day before extubation is used as the prediction window. The characteristic data two days before being separated from the breathing tube (and the second and third days before extubation) are used as the feature window (feature window). Therefore, the modeling data includes age, number of days using a respirator, and GCS, urine volume, injection volume, nutritional content, RASS, PIP, MAP, respiratory rate, and heart rate on the 2nd and 3rd days before extubation respectively.

Before analysis, the subject characteristic data collected in Example 1 can be subjected to data preprocessing procedures. The so-called data preprocessing includes removing abnormal data and inputting missing data. Among them, abnormal data refers to values that exceed the reasonable range of variables. In this example, the reasonable range of variables is set by the doctor. For example, each variable The reasonable range is as follows: age is 1-100 years old, number of days wearing a respirator is 1-60 days, GCS is 3-15, urine output is 0-5000ml, injection volume is 0-10000ml, and nutritional amount is 0-3000ml , RASS is -5-+4, PIP is 0-50, MAP is 10-40, respiratory rate is 0-40, heart rate is 0-300; and in this example, the average value of each variable is entered to fill in the gaps. data.

In addition, before conducting data analysis with the machine learning model, all data were standardized from +1 to -1; and, in order to avoid the occurrence of sampling bias, two sets of data were used in extubated subjects, one of which was The data of the day before extubation, the other group is random data, and five groups of data are randomly selected from the subjects who have not been extubated; the ratio of the characteristic data of the extubated subjects to the characteristic data of the subjects who have not been extubated is 1:3.4.

The analysis results of each machine learning model are shown in Figure 3 and Table 3. The accuracy in Table 3 is calculated according to the following formula: (TP+TN)/(TP+FN+TN+FP).

Table 3: Performance indicators of each machine learning model for extubation prediction

It can be seen from the results in Figure 3 and Table 3 that compared to the lower accuracy of LR, the accuracy of the other four machine learning models is high. Specifically, the AUCs of XGBoost, LightGBM, CatBoost and RF are 0.921 and 0.921 respectively. , 0.920 and 0.918; and from Figure 3B, it can be seen that there is good consistency between the predicted values of each machine learning model and the actual observed values, and XGBoost has the best consistency: and from the results of Figure 3C, it can be seen that all five machine learning models have It has certain clinical effectiveness, among which XGBoost and LightGBM perform best.

Example 3: Feature correlation analysis related to extubation prediction

In this example, the machine learning model used is XGBoost.

According to the four main clinical blocks revealed in Example 1, the 20 key features obtained in Example 2 were classified into the corresponding main clinical blocks, and the importance of each key feature was analyzed, and by accumulating the main clinical The importance value of the key features in the block is obtained to obtain the main clinical blocks and extubation The importance between predictions, the results are shown in Figure 4. From the results in Figure 4, it can be seen that the importance of consciousness/cognition block, body fluid balance block, respiratory function block, and physiological parameter block in intensive care are 0.284, 0.425, 0.232, and 0.045 respectively.

Then, we used SHAP values to determine how the 20 key features affect the possibility of extubation. The results are shown in Figure 5. It can be seen from the results in Figure 5 that improvement of GCS and increase in urine output are positively correlated with a higher probability of extubation one day later; while a high volume of injected fluid is negatively correlated with the probability of extubation.

Furthermore, as shown in Figure 6A to Figure 6F, through the partial dependence plot (PDP), we can obtain how key features such as GCS, RASS, urine volume, injection volume, PIP, and MAP affect the machine learning model evaluation respectively. Probability of extubation. The results from Figure 6A to Figure 6F can show that by visualizing each main clinical block or/and each characteristic, it can be used as a basis for explaining the success rate of extubation, that is, through each main clinical block or/and each characteristic The SHAP value or partial dependence diagram of the features can enable clinicians to know the reasons or the basis for the extubation prediction results obtained through the method of establishing extubation prediction using the machine learning model disclosed in this work.

Example 4: Analyze the impact of key features on prediction of extubation

Figures 7A and 7B illustrate the overall impact of key features on the prediction of extubation for two different individuals through the LIME and SHAP force values of key features respectively. In Figures 7A and 7B , red represents the total predicted probability of extubation. Variables that have an incremental effect (incremental effect), while blue represents variables that have a decremental effect (decremental effect) on the overall predicted probability of extubation. Specifically, it can be seen from the results in Figure 7A that although the injection volume of Case 1 on the second day before extubation was relatively high (2521ml), there are still many variables that are conducive to extubation, including clear consciousness (GCS of 14 and RASS is 0), high urine output (urine output on the 2nd day before extubation is 2450ml) and low respiratory rate (respiratory rate on the 2nd day before extubation is 14.5), therefore, the extubation prediction in Case 1 The probability is 0.81. On the contrary, it can be seen from the results in Figure 7B that Case 2 has many variables that are unfavorable to extubation, including high injection volume (2811ml one day before extubation) and high PIP (29.50cmH2O) and MAP (15.5mg/dL), so even if Case 2 has relatively clear consciousness (GCS is 15 and RASS is -1), the predicted probability of extubation in Case 2 is still only 0.19.

Example 5: The result of machine learning model calculation without key features

Referring to the content disclosed in Example 2, different machine learning models are used to analyze the characteristics of the subject's data. However, the difference is that the characteristics of the subject's data used to analyze the characteristics do not include the features of conscious blocks: GCS and RASS. Then, Check the precision, specificity, sensitivity, accuracy and AUROC of the results obtained by each machine learning model. The results are shown in Table 4 below.

Comparing the results in Table 3 and Table 4, it can be seen that when the key features disclosed in this creation are missing, the reliability of the predicted possibility of extubation will be reduced.

10: Establish an extubation prediction system using machine learning models

20:Database

30: Processing device

31:Data processing module

32: Model training module

33: Extubation prediction module

34:Visualization module

Claims (6)

A system for establishing extubation prediction using a machine learning model, which includes: a database that collects characteristic data of a plurality of patients, in which each patient has a record of using a respiratory assist device during hospitalization, and the characteristics The data includes a consciousness data, an input/output fluid data, a respiratory function data and a physiological parameter data; a processing device is connected to the database and has a data processing module to obtain the characteristics of a patient to be tested. Extract a key feature data from the data, wherein the key feature data includes the physiological parameter data, as well as the consciousness data and input obtained one day or/and two days before the date of extubation prediction. /Output the fluid data and the respiratory function data; an extubation prediction module analyzes the key characteristic data with an extubation prediction model to obtain the extubation of the patient to be tested within 24 hours after the date of extubation prediction. probability. Establish an extubation prediction system using a machine learning model as described in claim 1, wherein the extubation prediction module analyzes the key feature data using the extubation prediction model to generate the key feature data and the extubation probability. information related to each other. As described in claim 2, a system for predicting extubation is established using a machine learning model, wherein the processing device further includes a visualization module that converts the correlation information between the key characteristic data and the probability of extubation into a Visual interface. Establish a system for extubation prediction using a machine learning model as described in request 1, wherein the extubation prediction model is selected from XGBoost (Extreme Gradient Boosting), CatBoost (Categorical Boosting), LightGBM (Light Gradient Boosting Machine), Random Forest A group composed of algorithm (Random Forest, RF) and logistic regression (Logistic Regression, LR). As claimed in claim 1, a system for predicting extubation is established using a machine learning model, wherein the processing device further includes a model training module; The model training module trains a machine learning model using at least part of the characteristic data of the patients, and can perform verification to generate an extubation prediction model and a key feature. As described in request 5, a system for predicting extubation is established using a machine learning model, wherein the data processing module reads the characteristic data of the patients in the database and performs a data preprocessing procedure to delete data exceeding a certain threshold. Define a reasonable range of data or/and fill in missing parts of the data.