Emergency and critical patient early warning and screening system for detecting VOCs in exhaled air based on CI-TOFMS

Abstract

The invention discloses an acute and critical patient early warning screening system for detecting VOCs in exhaled air based on CI-TOFMS, and belongs to the field of early screening. Comprising a sample introduction processing system, a CI-TOFMS and a data processing module, the sample introduction processing system is used for collecting and preprocessing an expired gas sample, the CI-TOFMS is used for detecting the expired gas sample to obtain an eVOCs spectrogram, and the data processing module is used for collecting eVOCs spectrogram data and outputting an early warning result; the data processing module comprises a data input module, an eVOCs-based acute and critical symptom prediction model and a result output module. According to the system, the health state of the tested person can be evaluated only by collecting respiratory gas, and the system is very convenient. The established eVOCs-based acute and critical symptom prediction model further improves the accuracy and sensitivity of the eVOCs detection technology in the diagnosis of acute and critical patients, and provides a more reliable diagnosis basis for clinicians.

Description

Technical Field

[0001] The present invention belongs to early disease prediction technology and relates to an emergency and critical patient early warning screening system based on CI-TOFMS detection of exhaled breath VOCs. Background Art

[0002] Emergency medicine requires rapid assessment and judgment of acute and critical illnesses, followed by emergency treatment, and provides emergency medical rescue services for public health emergencies. Acute and critical illnesses (such as cardiac arrest, acute coronary syndrome, acute pulmonary embolism, aortic dissection, acute respiratory distress syndrome, sepsis, and acute pancreatitis) are characterized by suddenness, unpredictability, complexity, rapid progression, high mortality, and poor prognosis, ultimately leading to various types of shock and poor prognosis of vital organ failure. At the pathophysiological level, acute and critical illnesses generally have internal environmental disorders, metabolic disorders, and energy supply and utilization disorders, which often lead to impaired cell function. If not controlled in time, the condition will worsen further, eventually leading to multiple organ dysfunction, thus endangering the patient's life.

[0003] With the development of technology, blood and urine tests have brought great convenience to the diagnosis of diseases. However, due to the long detection time, the diagnosis of some patients may be delayed. VOCs in human exhaled breath are produced by cells through metabolic processes, and previous studies have identified the physiological links between VOCs and oxidative stress, lipid and amino acid metabolism, and inflammation. Studies have shown that human breath contains a complex mixture of hundreds of VOCs, and breath analysis has been used to diagnose, treat or monitor many diseases, such as chronic obstructive pulmonary disease, cancer, asthma, and diabetes, so VOCs can be used as biomarkers for overall health.

[0004] Current research on breath analysis either lacks specific identification of key VOCs species or lacks adequate analysis of multidimensional VOCs data through advanced algorithms such as machine learning to provide better classification performance. Detecting specific levels of VOCs in exhaled breath as biomarkers for critical illness has the potential to provide patients with a low-burden, rapid, and non-invasive tool. Summary of the invention

[0005] The problem to be solved by the present invention is to provide a critical patient early warning screening system based on chemical ionization time-of-flight mass spectrometry for detecting volatile organic compounds in exhaled breath.

[0006] The technical solution of the present invention is as follows:

[0007] An emergency and critical patient early warning screening system based on CI-TOFMS detection of exhaled breath VOCs, comprising a sample processing system, a CI-TOFMS, and a data processing module; the sample processing system is used to collect and pre-process exhaled breath samples, the CI-TOFMS is used to detect exhaled breath samples to obtain eVOCs spectra, and the data processing module is used to collect eVOCs spectrum data and output early warning results;

[0008] The data processing module includes a data input module, an emergency and critical illness prediction model based on eVOCs, and a result output module;

[0009] The method for constructing the critical illness prediction model based on eVOCs is as follows:

[0010] S1: Detect the exhaled breath samples of critically ill patients and healthy people of different degrees by chemical ionization time-of-flight mass spectrometry to obtain eVOCs spectra, and compare the different eVOCs spectra to obtain the eVOCs markers with significant differences;

[0011] S2: Collect the spectra data of eVOCs markers with significant differences, process the data, extract features, and obtain the eVOCs feature set;

[0012] S3: The eVOCs feature set, basic patient characteristics and clinical manifestation data are integrated and processed, and machine learning methods are used for training to establish an eVOCs-based critical illness prediction model.

[0013] Furthermore, the machine learning method includes at least one of a logistic regression model, an XGBoost model, a random forest model, a support vector model, a K nearest neighbor algorithm model, and a naive Bayes model.

[0014] Furthermore, in S2, the spectra of eVOCs markers with significant differences were analyzed by principal component analysis, and the mass spectrometry data were reduced in dimension, denoised, and standardized to obtain the eVOCs feature set.

[0015] Furthermore, in S3, the patient's basic characteristics and clinical manifestation data include age, gender, BMI, blood oxygen saturation, heart rate, and blood pressure.

[0016] Further, in S3, the eVOCs feature set, the basic characteristics of the patient and the clinical manifestation data are calculated by the following formula to obtain the target variable Rj, and the target variable Rj data set is trained using a machine learning method to establish an emergency and critical illness prediction model based on eVOCs;

[0017]

[0018] in,

[0019] Cj is the basic characteristics of the patient;

[0020] Sj is the patient's clinical data;

[0021] Vj is the eVOCs feature set data;

[0022] T response The time from the onset of symptoms to the patient's response;

[0023] T symptom It is the time when the symptoms of the disease appear, which is manifested by the signs perceived by the patient;

[0024] P VOCs is the rate of change of eVOCs concentration;

[0025] Lj represents the personalized information of the patient;

[0026] α1, α2, and α3 are scaling factors, which are determined by hyperparameter adjustment during model training;

[0027] δ is the disease type factor, and δ=1 before the disease is classified.

[0028] Furthermore, the Lj represents the personalized information of the patient, and is assigned by clinical medical staff according to abnormal fluctuations of eVOCs.

[0029] Furthermore, the chemical ionization time-of-flight mass spectrometry is a photoinduced chemical ionization time-of-flight mass spectrometry.

[0030] Furthermore, the light source of the photochemical ionization time-of-flight mass spectrometer is a radio frequency ultraviolet lamp with an energy of 10.6 eV, and the injection line temperature is set to 150 degrees.

[0031] Furthermore, the sampling processing system includes an exhaled air collection channel, a CO2 sensor arranged on the inner wall of the exhaled air collection channel, a sampling switch electrically connected to the CO2 sensor, and a sampling container connected to the exhaled air collection channel; the sampling switch is arranged at the connection point between the exhaled air collection channel and the sampling container, and is used to control the opening and closing of the sampling container.

[0032] Furthermore, it also includes S4: classifying the output results of the critical and severe disease prediction model based on eVOCs into critical and severe disease types, and standardizing them according to the disease mortality rate to obtain the disease type factor δ.

[0033] The present invention also provides an application of an emergency and critical patient early warning screening system based on CI-TOFMS detection of exhaled breath VOCs in predicting the progression rate of emergency and critical diseases and assisting in determining mortality and disease types.

[0034] Beneficial effects of the present invention:

[0035] 1) Easy to use. The system of the present invention only needs to collect exhaled breath samples from people to achieve health status assessment and determine the critical and severe status of patients. In addition, the model algorithm can be adjusted according to the type of disease. For example, in the application of cardiovascular diseases, the exhaled breath components include elevated acetone levels in the exhaled breath of patients with heart failure.

[0036] 2) Rapid detection. The collected samples can be detected in just 1 second, and the test results can be obtained within 1 minute, realizing rapid clinical detection.

[0037] 3) The test results are highly representative. The results of the present invention showed good diagnostic performance in both the training set and the test set, with the area under the ROC curve (AUC) exceeding 0.95. The present invention constructs an emergency and critical illness prediction model based on eVOCs, further improves the accuracy and sensitivity of eVOCs detection technology in the diagnosis of emergency and critical patients, and provides clinicians with a more reliable diagnostic basis. In addition, the application of volatile organic compound detection technology is not limited to the diagnosis of emergency and critical patients, but can also be extended to other fields. Combined with the detection of components in exhaled breath, it is also expected to be used for disease risk assessment and the formulation of personalized treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 Figure 3 is the relative impact of X46, X58, X64, X55 and X48 volatile organic compounds on the prediction model; A is a detailed view of the impact of each feature on a single prediction; B is the average absolute SHAP value of the volatile organic compounds; C to G are the relationship between the actual value of each volatile organic compound X46, X58, X64, X55 and X48 and its corresponding SHAP value; H is a heat map of the interaction between each volatile organic compound.

[0039] Figure 2 are the screening results of significantly different eVOCs; A is the volatile organic compounds with AUC values ​​greater than 0.9; B is the comparison of the Gini coefficients of ten key eVOCs; C is the five volatile organic compounds (X46, X58, X64, X55 and X48) determined by the VSOLassoBag algorithm; D is the distribution of the expression level of each VOC in the two study groups. DETAILED DESCRIPTION

[0040] The above contents of the present invention are further described in detail below through specific implementation methods in the form of embodiments. However, this should not be understood as the scope of the above subject matter of the present invention being limited to the following examples. All technologies implemented based on the above contents of the present invention belong to the scope of the present invention. Obviously, according to the above contents of the present invention, according to the common technical knowledge and customary means in this field, other various forms of modification, replacement and change can also be made without departing from the above basic technical ideas of the present invention.

[0041] As a non-invasive technology, breath detection of exhaled volatile organic compounds (eVOCs) has potential value in disease diagnosis. In some embodiments of the present invention, a prospective research method was adopted to continuously select 121 critically ill patients who visited the emergency department of our hospital from January to March 2024, and 728 healthy volunteers were used as controls. The independently developed chemical ionization time-of-flight mass spectrometry (CI-TOFMS) platform was used to detect volatile organic compounds (VOCs) in exhaled gas. The receiver operating characteristic (ROC) curve, ensemble learning algorithm and LASSO regression were used to screen key VOC biomarkers. Based on the key VOC biomarkers, four machine learning algorithms and formulas introducing different disease variables were used to construct a diagnostic model.

[0042] In some embodiments of the present invention, patients were divided into acute myocardial infarction subgroup and pneumonia subgroup according to the initial diagnosis, and the propensity score matching method was used to select controls to evaluate the performance of the model in a specific disease population. Results Based on the VOC detection results of healthy controls, a reference range of exhaled VOCs in healthy people was established, and a standard curve of 9 aldehyde compounds was drawn. Through a variety of statistical methods, 5 VOCs (X46, X58, X64, X55 and X48) were identified to have the best diagnostic performance in distinguishing critically ill patients from healthy controls. The 4 diagnostic models constructed based on these 5 biomarkers showed good diagnostic performance in both the training set and the test set, and the area under the ROC curve (AUC) exceeded 0.95. Conclusion This study used CI-TOFMS technology to detect eVOCs spectra and preliminarily explored its application potential in the rapid screening and evaluation of critically ill patients. This non-invasive diagnostic strategy is simple to operate and can improve the diagnostic efficiency of critically ill patients.

[0043] Example 1

[0044] The critical patient early warning screening system of the present invention:

[0045] The present invention provides a critically ill patient early warning screening system based on CI-TOFMS detection of exhaled breath VOCs, comprising a sample processing system, a CI-TOFMS, and a data processing module; the sample processing system is used to collect and pre-process exhaled breath samples, the CI-TOFMS is used to detect exhaled breath samples to obtain eVOCs spectra, and the data processing module is used to collect eVOCs spectra data and output early warning results;

[0046] Through the sample introduction and processing system, the patient's breath sample enters the photochemical ionization time-of-flight mass spectrometer to detect the volatile organic compounds (eVOCs) in the exhaled breath and obtain the eVOCs composition data of different samples.

[0047] The data processing module includes a data input module and a result output module for transmitting the VOCs composition data of different samples to the critical illness prediction model based on eVOCs.

[0048] The data input module obtains the eVOCs composition data of different samples from the sample detection module, and distinguishes the most discriminative eVOCs, i.e., the eVOCs feature set, between critically ill patients and healthy controls through the principal component analysis model (PCA), t-distributed stochastic neighbor embedding model (t-SNE), uniform manifold approximation and projection model (UMAP), integrated learning algorithm XGBoost model, and VSOLassoBag algorithm.

[0049] The target variable Rj is obtained by calculating the eVOCs feature set, the basic characteristics of the patient and the clinical manifestation data through the following formula. The target variable Rj data set is trained using the XGBoost machine learning method to establish an emergency and critical illness prediction model based on eVOCs (XGBoost model);

[0050]

[0051] in,

[0052] Cj is the basic characteristics of the patient;

[0053] Sj is the patient's clinical data;

[0054] Vj is the eVOCs feature set data;

[0055] T response The time from the onset of symptoms to the patient's response;

[0056] T symptom It is the time when the symptoms of the disease appear, which is manifested by the signs perceived by the patient;

[0057] P VOCs is the rate of change of eVOCs concentration;

[0058] Lj represents the personalized information of the patient;

[0059] α1, α2, and α3 are scaling factors, which are determined by hyperparameter adjustment during model training;

[0060] δ is the disease type factor, and δ=1 before the disease is classified.

[0061] The result output module is responsible for using the trained XGBoost model to perform predictive analysis on eVOCs samples in healthy status locations to obtain the detection results of the samples.

[0062] The results of the present invention showed good diagnostic performance in both the training set and the test set, with the area under the ROC curve (AUC) exceeding 0.95.

[0063] Example 2

[0064] Use eVOCs samples for early warning screening and analysis of critically ill patients.

[0065] 1. Methods

[0066] There were 121 critically ill patients in the group, all of whom were from Qilu Hospital of Shandong University. There were 728 healthy controls, all of whom were recruited volunteers. The demographic and pathological characteristics of the healthy controls and critically ill patients are shown in Table 1.

[0067]

[0068]

[0069] This study was approved by the Ethics Committee of Qilu Hospital of Shandong University and registered at ClinicalTrials.gov (NCT06277895). It followed the Standards for Reporting Diagnostic Accuracy (STARD) reporting guidelines. All participants signed written informed consent.

[0070] Healthy individuals in the control group were recruited from the Physical Examination Center of Qilu Hospital of Shandong University. Inclusion criteria included: 1) aged 18 years or older, 2) no history of tumor, no signs of liver and kidney dysfunction, no signs of infection, 3) no consumption of spicy and irritating foods (such as coffee, wine, garlic, etc.) in the past 24 hours, and 4) comprehensive clinical data. Exclusion criteria included: 1) history of cancer, 2) signs of active infection, or 3) liver or kidney dysfunction, 4) incomplete clinical and / or accessory examination records.

[0071] Patients in the experimental group were recruited from the emergency department of Qilu Hospital of Shandong University. Inclusion criteria included: 1) aged 18 years or older, 2) no cardiovascular or pulmonary disease, no liver or kidney dysfunction, no acute infection or pneumonia but no cardiovascular or pulmonary disease. Data collection in this study was carried out in accordance with local guidelines and indicators, and human data collection was approved by the Ethics Committee of Qilu Hospital of Shandong University (approval number: KYLL-202401-047).

[0072] We set standard sampling requirements and protocols, and trained investigators collected exhaled breath samples. For healthy and conscious individuals: 1) All participants need to prepare for sampling in advance: no smoking, drinking, or eating spicy and irritating foods (such as coffee, garlic, etc.) within 24 hours before sampling, 2) rinse the mouth with purified water immediately before sampling, 3) all samples are collected at the same location to reduce the impact of environmental factors. The subject takes a deep breath and inhales the air completely into the sampling bag through a disposable nozzle. The breath sampling bag with a volume of 1L is made of polytetrafluoroethylene (PTEF).

[0073] For intubated individuals: Connect the inlet of the gas collection bag to the port of the endotracheal tube. Observe the ventilator waveform interface, open the gas collection bag valve during exhalation, close the inlet valve immediately at the end of exhalation, and reconnect the ventilator circuit.

[0074] The chemical ionization time-of-flight mass spectrometer (CI-TOFMS) of this embodiment is mainly composed of a chemical ionization source, an ion transfer system, a vacuum system and a TOF mass spectrometer. The chemical ionization source is mainly composed of a VUV krypton lamp, a repelling electrode (V1), an injection electrode (V2), a segmented quadrupole (SQ) and a differential vacuum hole (Skimmer1). The exhaled breath sample is directly introduced into the ionization region through a PEEK capillary with an inner diameter of 0.5 mm, an outer diameter of 1 / 16 inch and a length of 1.2 meters in the photoinduced chemical ionization time-of-flight mass spectrometer. In order to eliminate the condensation of eVOCs and minimize possible surface adsorption, the PEEK capillary is heated to 150 degrees. The extraction cycle of ions is 40μs (25kHz), and the time resolution of a spectral data is 1s.

[0075] The 2.5th to 97.5th percentile interval of each compound established the medical reference range for 132 volatile organic compounds (eVOCs) detected in healthy controls. The obtained eVOCs sample data of critically ill patients and healthy people were first subjected to principal component analysis (PCA), t-distributed stochastic neighbor embedding (t-SNE) and uniform manifold approximation and projection (UMAP) to analyze the overall classification trend of VOCs expression profiles between groups.

[0076] The diagnostic performance of each VOC was evaluated using receiver operating characteristic (ROC) curve analysis. The area under the curve (AUC) of each eVOCs was calculated, and VOCs with AUC greater than 0.9 were selected for further analysis to obtain the eVOCs feature set. The ensemble learning algorithm XGBoost model was used to rank the importance of the detected VOCs. The intersection of the VOCs selected by the above two methods was subjected to feature selection by the VSOLassoBag algorithm. The final VOCs selected by VSOLassoBag were considered to be the key VOCs for distinguishing critically ill patients from healthy controls, and the target variable Rj was extracted for the key VOCs.

[0077]

[0078] Cj is the patient’s age, gender, BMI, blood oxygen saturation, heart rate, and blood pressure;

[0079] Sj is the patient’s age, sex, BMI, blood oxygen saturation, heart rate, and blood pressure;

[0080] Vj is the eVOCs feature set data;

[0081] T response The time from the onset of symptoms to the patient's response;

[0082] T symptom It is the time when the symptoms of the disease appear, which is manifested by the signs perceived by the patient;

[0083] P VOCs is the rate of change of eVOCs concentration;

[0084] Lj represents the personalized information of the patient;

[0085] The eVOCs feature set, basic patient characteristics and clinical manifestation data were further calculated using the following formula to obtain the target variable Rj. The target variable Rj data set was trained using a machine learning method to establish an emergency and critical illness prediction model based on eVOCs.

[0086] By detecting 112 volatile organic compounds that showed differences between healthy individuals and critically ill patients and with FDR-adjusted p values ​​< 0.05, this example calculated the AUC to evaluate their individual diagnostic performance. Among them, the AUC values ​​of 13 volatile organic compounds were greater than 0.9 ( Figure 2 A). We then used an ensemble machine learning strategy to evaluate the importance of these 112 VOCs in building a random forest model to distinguish between case and control groups. Based on the average reduction in the Gini coefficient, we identified the top ten most important VOCs ( Figure 2 B). The intersection of these methods resulted in 10 key VOCs: X46, X58, X64, X55, X48, X158, X50, X44, X63, and X56, representing small VOCs with molecular weights of 46, 58, 64, 55, 48, 158, 50, 44, 63, and 56 in exhaled breath, respectively, which can be used to characterize the differences between critically ill patients and healthy subjects. Subsequently, we applied the VSOLassoBag algorithm with 100 sampling iterations for feature selection to enhance robustness. The algorithm identified five VOCs (X46, X58, X64, X55, and X48) as the best markers for the final diagnostic model ( Figure 2C). By drawing a standard curve, these organic compounds were identified as acetaldehyde (46), methyl mercaptan (48), 1,3-butadiene (55), acrolein (55), n-pentene (55), cyclopentane (55), cis-trans, 2, pentene (55), n-butyraldehyde (55), propionaldehyde (58), butane (58), and 1,2-dichloroethane. The violin curve plots illustrate the distribution of the expression levels of each eVOCs in the two study groups ( Figure 2 D).

[0087] During the model building and evaluation process, the data were divided into training and validation sets in a ratio of 7:3. Six models were constructed, including logistic regression, XGBoost, random forest, support vector machine, K nearest neighbor algorithm model, and naive Bayes model to verify the correctness and classification performance of candidate biomarkers.

[0088] The SHAP method was used to clarify the output of our XGBoost model optimized with Rj by quantifying the contribution of each variable to the prediction, thereby enhancing the interpretability of the model. Figure 1 A provides a detailed view of the impact of each feature on a single prediction. Each point represents a prediction, the color indicates the feature value, and the position on the x-axis indicates the SHAP value. Figure 1 B shows the average absolute SHAP values ​​of these VOCs, arranged in descending order of their contribution. The results show that X46 has the most significant impact on the model output, while X55 has the least impact. To further clarify the impact of individual features on model predictions, Figure 1 C to 1G present the SHAP dependency plots, illustrating the relationship between the actual value of each biomarker and its corresponding SHAP value, thus highlighting the contribution of each biomarker to the prediction. For example, subjects with X46>2247 or X48>17307.6 have a SHAP value greater than zero, which pushes the decision toward the “Case” class. In addition, the interaction heatmap ( Figure 1 H) shows that in particular the main effects of the individual features X46 and X48 (see diagonal lines) are significantly higher than their interaction effects, indicating that these features mainly affect the model independently.

[0089] The results showed that eVOCs characteristic values ​​could be used to detect critically ill patients and healthy people with high accuracy, sensitivity and specificity.

[0090] In summary, the early warning screening system of the present invention can realize early warning screening for critically ill patients.

Claims

1. An early warning screening system for critically ill patients based on CI-TOFMS detection of exhaled VOCs, characterized by: It includes a sample processing system, a CI-TOFMS, and a data processing module; the sample processing system is used to collect and pre-process exhaled breath samples, the CI-TOFMS is used to detect exhaled breath samples to obtain eVOCs spectra, and the data processing module is used to collect eVOCs spectra data and output warning results; The data processing module includes a data input module, an emergency and critical illness prediction model based on eVOCs, and a result output module; The method for constructing the critical illness prediction model based on eVOCs is as follows: S1: Detect the exhaled breath samples of critically ill patients and healthy people of different degrees by chemical ionization time-of-flight mass spectrometry to obtain eVOCs spectra, and compare the different eVOCs spectra to obtain the eVOCs markers with significant differences; S2: Collect the spectra data of eVOCs markers with significant differences, process the data, extract features, and obtain the eVOCs feature set; S3: The eVOCs feature set, basic patient characteristics and clinical manifestation data are integrated and processed, and machine learning methods are used for training to establish an eVOCs-based critical illness prediction model.

2. The early warning screening system for critically ill patients based on CI-TOFMS detection of exhaled VOCs according to claim 1, characterized in that: The machine learning method includes at least one of a logistic regression model, an XGBoost model, a random forest model, a support vector model, a K nearest neighbor algorithm model, and a naive Bayes model.

3. The early warning screening system for critically ill patients based on CI-TOFMS detection of exhaled VOCs according to claim 1, characterized in that: In S2, the spectra of eVOCs markers with significant differences were analyzed by principal component analysis, and the mass spectrometry data were reduced in dimension, denoised, and standardized to obtain the eVOCs feature set.

4. The early warning screening system for critically ill patients based on CI-TOFMS detection of exhaled VOCs according to claim 1, characterized in that: In S3, the patient's basic characteristics and clinical manifestation data include age, gender, BMI, blood oxygen saturation, heart rate, and blood pressure.

5. The early warning screening system for critically ill patients based on CI-TOFMS detection of exhaled VOCs as claimed in claim 4, characterized in that: In S3, the eVOCs feature set, the basic characteristics of the patient and the clinical manifestation data are calculated by the following formula to obtain the target variable Rj. The target variable Rj data set is trained using a machine learning method to establish an emergency and critical illness prediction model based on eVOCs; in, Cj is the basic characteristics of the patient; Sj is the patient's clinical data; Vj is the eVOCs feature set data; T response The time from the onset of symptoms to the patient's response; T symptom It is the time when the symptoms of the disease appear, which is manifested by the signs perceived by the patient; P VOCs is the rate of change of eVOCs concentration; Lj represents the personalized information of the patient; α1, α2, and α3 are scaling factors, which are determined by hyperparameter adjustment during model training; δ is the disease type factor, and δ=1 before the disease is classified.

6. The critically ill patient early warning screening system based on CI-TOFMS detection of exhaled breath VOCs according to claim 1, characterized in that: The chemical ionization time-of-flight mass spectrometry is a photoinduced chemical ionization time-of-flight mass spectrometry.

7. The early warning screening system for critically ill patients based on CI-TOFMS detection of exhaled VOCs according to claim 5, characterized in that: The light source of the photochemical ionization time-of-flight mass spectrometer is a radio frequency ultraviolet lamp with an energy of 10.6 eV, and the temperature of the injection line is set to 150 degrees.

8. According to the early warning screening system for critically ill patients based on CI-TOFMS detection of exhaled breath VOCs as described in claim 1, the sampling processing system includes an exhaled breath collection channel, a CO2 sensor arranged on the inner wall of the exhaled breath collection channel, a sampling switch electrically connected to the CO2 sensor, and a sampling container connected to the exhaled breath collection channel; the sampling switch is arranged at the connection point between the exhaled breath collection channel and the sampling container, and is used to control the opening and closing of the sampling container.

9. The early warning screening system for critically ill patients based on CI-TOFMS detection of exhaled breath VOCs according to claim 5, characterized in that: It also includes S4: classifying the output results of the critical and severe disease prediction model based on eVOCs into critical and severe disease types, standardizing them according to the disease mortality rate, and obtaining the disease type factor δ.

10. The early warning screening system for critically ill patients based on CI-TOFMS detection of exhaled VOCs according to claim 5, characterized in that: The Lj represents the patient's personalized information, which is assigned by clinical medical staff according to abnormal eVOCs fluctuations.

11. Application of the early warning screening system for critically ill patients based on CI-TOFMS detection of exhaled breath VOCs as described in claim 1 in predicting the progression rate of critically ill diseases and assisting in determining mortality and disease types.

Images