Method and system for constructing chronic obstructive pulmonary screening model based on forced oscillation lung function

Abstract

The invention discloses a method for establishing a chronic obstructive pulmonary disease screening model based on a forced oscillation lung function, belongs to the technical field of clinical diagnostics, and aims to solve the problem that screening of chronic obstructive pulmonary disease by using a spirometer needs to consume a large amount of time and economic cost. 2, dividing the collected clinical data into a training set and a test set for constructing an FEV1 / FVC prediction model; 3, constructing an FEV1 / FVC prediction model based on an FOT (Forced Oscillation Test) parameter and a basic demographic parameter; 4, searching an optimal boundary value of the prediction model; 5, evaluating the screening performance of the prediction model on the chronic obstructive pulmonary disease in the test set; and 6, evaluating the screening performance of the prediction model on the chronic obstructive pulmonary disease in an external verification set. According to the method, the chronic obstructive pulmonary disease screening model is used for preliminarily screening people needing to be subjected to spirometer examination, early diagnosis and early treatment of the chronic obstructive pulmonary disease are facilitated, and the social and economic burden caused by the chronic obstructive pulmonary disease is relieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of clinical diagnostics, and in particular relates to a method and system for constructing a COPD screening model based on forced oscillation lung function. Background Art

[0002] Chronic obstructive pulmonary disease (COPD) is abbreviated as COPD, which has a high prevalence and mortality rate. The Global Burden of Disease Study shows that the global prevalence of COPD reached 212.3 million in 2019, resulting in 3.3 million deaths. The prevalence of COPD in adults over 20 years old in my country is 8.6%, and the prevalence of COPD in adults over 40 years old is 13.7%. COPD has caused serious disease burden and socioeconomic impact.

[0003] Spirometry is the gold standard for diagnosing COPD, but this examination process requires a high degree of respiratory coordination from the subject and requires strict and forceful breathing movements, which is difficult for children, the elderly or people with weak constitutions to complete smoothly. In my country, many primary care workers do not have enough time and standardized technology to complete spirometry examinations with strict quality control requirements. In addition, large-scale spirometry examinations in the general population to screen for COPD require a lot of time and economic costs, so this examination is difficult to be widely promoted in primary medical institutions.

[0004] In the past, some simple COPD screening tools have been developed to identify people at high risk of COPD, and then perform spirometry on high-risk people to improve the diagnostic efficiency of COPD. Currently, COPD screening tools mainly include two categories, one is screening questionnaires, and the other is portable physical measurement devices. However, these tools still have some shortcomings in practical applications: questionnaires are not sensitive enough for early COPD patients with mild symptoms; portable physical measurement devices such as COPD-6 and peak flow meters require subjects to breathe hard, which may be difficult or inaccurate for older people and those who do not cooperate well.

[0005] Forced Oscillation Technique (FOT) is a non-invasive method that does not rely on forced exhalation to measure respiratory mechanical characteristics such as airway resistance. Its principle is to apply an oscillating wave (pressure signal) to the respiratory system during quiet breathing, and measure the corresponding flow changes generated by the respiratory system under the oscillating pressure signal to calculate the impedance of the respiratory system. Among them, the FOT measurement using a rectangular pulse signal and Fourier transformation into a sine wave of different frequencies is the impulse oscillation (IOS) examination. FOT and IOS examinations require low cooperation from the subjects, and only quiet breathing is required to complete the examination. The examination is convenient and easy, and is applicable to a wide range of people. It has good application prospects in primary medical institutions. At present, it is necessary to establish a COPD screening method based on forced oscillation lung function to conduct preliminary screening of people who need to undergo spirometry examination, which is conducive to improving the diagnostic efficiency of COPD in primary medical units, promoting early diagnosis and treatment of COPD, and reducing the socioeconomic burden caused by COPD. Summary of the invention

[0006] In view of the above problems, the present invention establishes a method and system for establishing a COPD screening model based on forced oscillation lung function, which can conduct preliminary screening of people who need to undergo spirometry examination, thereby facilitating early diagnosis and treatment of COPD and reducing the socioeconomic burden caused by COPD.

[0007] One object of the present invention is to provide a method for establishing a COPD screening model based on forced oscillation lung function, the method comprising the following steps:

[0008] Step 1: Collect the spirometry and FOT examination data of the subjects on the same day and establish a clinical data set; the clinical data includes spirometry parameters, FOT parameters and basic demographic information;

[0009] Step 2: Divide the collected clinical data into a training set and a test set, with the ratio of the training set to the test set being 7:3 or 1:1; the training set is used to build the FEV1 / FVC prediction model, and the test set is used to verify the discrimination and calibration of the model;

[0010] Step 3: Using the spirometry parameter FEV1 / FVC as the outcome variable, and the FOT parameter and demographic data as the predictor variables, the training set was subjected to multivariate linear regression, the regression coefficient β value of each variable was calculated, and the FEV1 / FVC prediction model based on the FOT parameter and basic demographic parameters was constructed, and the calculation formula was generated;

[0011] Step 4: Calculate the predicted FEV1 / FVC value according to the prediction formula, calculate the sensitivity and specificity of the FEV1 / FVC prediction model for identifying COPD at different cutoff values, and find the optimal cutoff value of the model;

[0012] Step 5: Conduct internal validation of the FEV1 / FVC prediction model in the development cohort and evaluate the model's screening performance for COPD in the test set;

[0013] Step 6: Conduct external validation of the FEV1 / FVC prediction model in other centers to evaluate the model's screening performance for COPD in the external validation set.

[0014] Furthermore, the basic demographic information described in step 1 includes gender, age, height, and weight, and the FOT parameters include resistance at 5 Hz, resistance at 20 Hz, resonance frequency, reactance at 5 Hz, respiratory impedance at 5 Hz, and reactance area.

[0015] Furthermore, in step 3, the calculation formula of the FEV1 / FVC prediction model is as follows:

[0016] FEV1 / FVC=constant term+β1×age+β2×sex+β3×height+β4×weight+β5×R5+β6×R20+β7×X5+β8×Z5+β9×Fres+β 10 ×AX

[0017] In the above, β1 represents the regression coefficient of age, β2 represents the regression coefficient of gender, β3 represents the regression coefficient of height, β4 represents the regression coefficient of weight, β5 represents the regression coefficient of R5, R5 represents the resistance at 5 Hz, β6 represents the regression coefficient of R20, R20 represents the resistance at 20 Hz, β7 represents the regression coefficient of X5, X5 represents the reactance at 5 Hz, β8 represents the regression coefficient of Z5, Z5 represents the respiratory impedance at 5 Hz, β9 represents the regression coefficient of Fres, Fres represents the resonance frequency, β 10 represents the regression coefficient of AX, where AX represents the reactance area.

[0018] All the parameters mentioned above can be derived from forced oscillation pulmonary function tests or from impulse oscillation (IOS) tests that use rectangular forced signals and transform them into sine waves of different frequencies via Fourier transform.

[0019] The number of subjects required for modeling must be >500, because there may be multicollinearity between FOT parameters. If the modeling sample size is insufficient, it may lead to overfitting of the model. In order to avoid overfitting of the model, it is recommended to use machine learning algorithms including stepwise regression / optimal subsets / LASSO / ridge regression / elastic network method for modeling, and select the method with the best model performance in the test set to establish the final model.

[0020] Furthermore, the diagnostic criteria for COPD in step 4 are recommended to be FEV1 / FVC < 0.7 (airflow obstruction) after inhalation of bronchodilators. If there is no condition for bronchodilator test, FEV1 / FVC < 0.7 before inhalation of bronchodilators can be used. Taking the diagnosis of COPD as the gold standard, the model-estimated FEV1 / FVC prediction value is used to calculate the area under the ROC curve AUC value for screening pulmonary function-defined COPD, and the sensitivity, specificity, Youden index, accuracy, positive predictive value, negative predictive value, positive likelihood ratio and negative likelihood ratio of the model for COPD screening at different FEV1 / FVC prediction value cutoffs are calculated to determine the optimal classification value of the model for screening pulmonary function-defined COPD:

[0021] The single cut-off value model was used to calculate the Youden index at different cut-off values ​​of the FEV1 / FVC prediction model. Youden index = sensitivity + specificity - 1. The cut-off value when the Youden index was the largest was the optimal cut-off value.

[0022] Or adopt a double cut-off value model, define sensitivity > 90% as the exclusion cut-off value, specificity > 90% as the inclusion cut-off value, and the gray area between the inclusion cut-off value and the exclusion cut-off value;

[0023] Or a double cut-off value model is used, with negative predictive value > 90% defined as the exclusion cut-off value, positive predictive value > 90% defined as the inclusion cut-off value, and the gray area between the inclusion cut-off value and the exclusion cut-off value;

[0024] Or a double-cutoff model is used, with the cutoff value when the Youden index is the largest as the optimal cutoff value, to identify COPD (defined as FEV1 / FVC < 0.7 after inhalation of bronchodilator) as the model inclusion cutoff value, and to identify airflow obstruction (defined as FEV1 / FVC < 0.7 before inhalation of bronchodilator) as the model exclusion cutoff value. This is to include as many COPD patients as possible and exclude as many patients without airflow obstruction as possible.

[0025] Furthermore, when the double-cutoff value model is used in step 4, the standard for judging whether the subject should undergo spirometry examination using the predicted FEV1 / FVC value calculated by the FEV1 / FVC prediction model is:

[0026] If the predicted FEV1 / FVC value falls within the inclusion threshold, the subject is judged to have COPD; if the predicted value falls within the exclusion threshold, the subject is judged to be a healthy person; if the predicted value falls into the gray area, the subject is judged to have suspected COPD, and a spirometry examination is required to confirm the diagnosis.

[0027] Furthermore, the specific operations of step 5 for internal validation in the FEV1 / FVC prediction model development cohort are as follows:

[0028] Step 501: Calculate the predicted FEV1 / FVC values ​​of the subjects in the training set and the test set using the calculation formula of the FEV1 / FVC prediction model;

[0029] Step 502: Calculate the mean absolute error (MAE) and root mean square error (RMSE) between the predicted FEV1 / FVC value and the actual measured FEV1 / FVC value, take COPD defined by the diagnosis of lung function as the prediction result, draw the ROC curve of the FEV1 / FVC prediction model in the training set and the test set, and calculate the area under the ROC curve AUC value;

[0030] Step 503: Draw a calibration curve of the FEV1 / FVC prediction model, presenting the deviation between the predicted FEV1 / FVC value and the actual measured FEV1 / FVC value in the training set and the test set. The smaller the deviation, the higher the accuracy of the model.

[0031] Furthermore, in step 6, an external data set is used to perform external validation on the constructed FEV1 / FVC prediction model, including the following steps:

[0032] Step 601: sorting the external data set according to the data set structure for constructing the FEV1 / FVC prediction model;

[0033] Step 602: Calculate the mean absolute error (MAE) and root mean square error (RMSE) of the FEV1 / FVC prediction model in the external data set, take the measured FEV1 / FVC value <0.7 as the diagnostic standard for COPD, calculate the area under the ROC curve AUC value of the FEV1 / FVC prediction model for screening COPD in the external data set, and calculate the sensitivity, specificity, accuracy, positive predictive value, negative predictive value, positive likelihood ratio and negative likelihood ratio of the FEV1 / FVC prediction model.

[0034] Another object of the present invention is to provide a system for constructing a COPD screening model based on forced oscillation lung function, the system comprising:

[0035] A data collection module is used to collect the spirometry and FOT examination data of the subjects on the same day and establish a clinical data set; wherein the clinical data includes spirometry parameters, FOT parameters and basic demographic information;

[0036] Data processing, used to randomly allocate clinical data into training and test sets, with the ratio of training set to test set being 7:3 or 1:1;

[0037] The model building module uses the spirometry parameters FEV1 / FVC in the training set as the outcome variable, the FOT parameters and demographic data as the predictor variables, performs multivariate linear regression on the training set, calculates the regression coefficient β value of each predictor variable, builds the FEV1 / FVC prediction model based on the FOT parameters and basic demographic parameters, and generates the calculation formula;

[0038] Model optimization module, used to calculate the predicted FEV1 / FVC value according to the FEV1 / FVC prediction model formula, and use FEV1 / FVC < 0.7 as the diagnostic standard for COPD, calculate the sensitivity and specificity of the model for identifying COPD at different cut-off values, and find the optimal cut-off value of the model;

[0039] The validation module is used to perform internal validation of the FEV1 / FVC prediction model in the development cohort using the validation set, and is also used to perform external validation of the FEV1 / FVC prediction model using an external dataset.

[0040] Furthermore, the calculation formula of the FEV1 / FVC prediction model constructed in the model building module is as follows:

[0041] FEV1 / FVC=constant term+β1×age+β2×sex+β3×height+β4×weight+β5×R5+β6×R20+β7×X5+β8×Z5+β9×Fres+β 10 ×AX

[0042] In the above, β1 represents the regression coefficient of age, β2 represents the regression coefficient of gender, β3 represents the regression coefficient of height, β4 represents the regression coefficient of weight, β5 represents the regression coefficient of R5, R5 represents the resistance at 5 Hz, β6 represents the regression coefficient of R20, R20 represents the resistance at 20 Hz, β7 represents the regression coefficient of X5, X5 represents the reactance at 5 Hz, β8 represents the regression coefficient of Z5, Z5 represents the respiratory impedance at 5 Hz, β9 represents the regression coefficient of Fres, Fres represents the resonance frequency, β 10 represents the regression coefficient of AX, where AX represents the reactance area.

[0043] All the parameters mentioned above can be derived from forced oscillation pulmonary function tests or from impulse oscillation (IOS) tests that use rectangular forced signals and transform them into sine waves of different frequencies via Fourier transform.

[0044] Compared with the prior art, the present invention has the following advantages:

[0045] 1. FEV1 / FVC < 0.7 measured by spirometry is the gold standard for the diagnosis of COPD, but the test subject needs to breathe hard during the spirometry test, the test time is long, and the quality control requirements are high, which makes it difficult to promote at the grassroots level. Compared with the spirometry test, the FOT test in this application only requires the subject to breathe calmly, the test speed is faster, the requirements for the subject's cooperation are lower, it is easier to complete, and the test can be completed using portable equipment at grassroots community health service centers. It is conducive to the rapid and convenient screening of potential COPD patients in grassroots medical institutions, promoting the early diagnosis and early treatment of COPD, and reducing the socioeconomic burden caused by COPD.

[0046] 2. Currently, there are two main types of COPD screening tools: screening questionnaires and portable physical measurement devices. Questionnaires are not sensitive enough for patients with early COPD with mild symptoms; portable physical measurement devices such as COPD-6 and peak flow meters require subjects to breathe hard, which may be difficult or inaccurate for older people with poor cooperation. We used the spirometer parameter FEV1 / FVC as the outcome indicator, FOT parameters and demographic data as predictor variables, and established a FEV1 / FVC prediction model based on FOT, which can estimate the FEV1 / FVC of the subjects more directly and objectively, and use the FEV1 / FVC estimated by the model to screen COPD in the population, which is conducive to improving the accuracy of COPD screening. In addition, our model can be combined with screening questionnaires or peak flow meters to further improve the accuracy of COPD screening, which will help change the current clinical pathway for early screening and diagnosis of COPD, greatly improve the diagnostic efficiency of COPD in primary medical institutions, and significantly improve the early diagnosis level of COPD. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 Schematic diagram of the process of establishing a COPD screening model based on forced oscillation lung function;

[0048] Figure 2 Research flow chart in the embodiment;

[0049] Figure 3 ROC curve of the COPD screening model based on forced oscillation lung function for identifying airflow obstruction in the development cohort. (A) ROC curve of the model for identifying airflow obstruction in the training set; (B) ROC curve of the model for identifying airflow obstruction in the test set.

[0050] Figure 4 Scatter plot of correlation between predicted values ​​and actual values ​​of COPD screening model based on forced oscillation lung function, (A) Scatter plot of correlation between predicted values ​​and actual values ​​of the model in the training set; (B) Scatter plot of correlation between predicted values ​​and actual values ​​of the model in the test set;

[0051] Figure 5ROC curve of the COPD screening model based on forced oscillation lung function for identifying COPD in the development cohort;

[0052] Figure 6 Flow chart of the forced oscillation lung function model for screening COPD in clinical practice;

[0053] Figure 7 The discrimination and calibration of the model in the external validation set, (A) scatter plot of the correlation between the predicted value and the actual value of the model in the external validation set; (B) ROC curve of the model in identifying airflow obstruction in the external validation set; (C) ROC curve of the model in identifying COPD in the external validation set. DETAILED DESCRIPTION

[0054] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more definite definition of the protection scope of the present invention.

[0055] like Figure 1 As shown, this embodiment provides a method for establishing a COPD screening model based on forced oscillation lung function, and the establishment method includes the following steps:

[0056] Step 1: Collect the spirometry and FOT examination data of the subjects on the same day and establish a clinical data set; the clinical data includes spirometry parameters, FOT parameters and basic demographic information;

[0057] Step 2: Divide the collected clinical data into a training set and a test set. The training set is used to build the FEV1 / FVC prediction model, and the test set is used to verify the discrimination and calibration of the model.

[0058] Step 3: Using spirometry parameters as outcome variables, FOT parameters and demographic data as predictor variables, multivariate linear regression was performed on the training set, the regression coefficient β value of each variable was calculated, and a FEV1 / FVC prediction model based on FOT parameters and basic demographic parameters was constructed, and a calculation formula was generated;

[0059] Step 4: Calculate the predicted FEV1 / FVC value according to the prediction formula, calculate the sensitivity and specificity of the FEV1 / FVC prediction model for identifying COPD at different cutoff values, and find the optimal cutoff value of the model;

[0060] Step 5: Conduct internal validation in the FEV1 / FVC prediction model development cohort to evaluate the model’s screening performance for COPD in the test set;

[0061] Step 6: Conduct external validation of the FEV1 / FVC prediction model in other centers to evaluate the model's screening performance for COPD in the external validation set.

[0062] Specifically, in step 1, demographic information and lung function test data of COPD patients and healthy subjects are collected prospectively or retrospectively:

[0063] COPD patients were defined as follows: (1) aged ≥ 20 years; (2) lung function met the diagnostic criteria for COPD, i.e. forced expiratory volume in one second (FEV1) / forced vital capacity (FVC), i.e. FEV1 / FVC < 0.7. Healthy subjects were defined as subjects with lung function FEV1 / FVC ≥ 0.7 and no history of chronic respiratory diseases.

[0064] The fields in the data set in step 1 include the basic information of the subjects: date of birth, age, gender, height, weight, smoking history, and date of pulmonary function test; the spirometry data include: FEV1, FVC, and FEV1 / FVC; the forced oscillation pulmonary function data include: resistance at 5Hz (R5), resistance at 20Hz (R20), resonant frequency (Fres), reactance at 5Hz (X5), respiratory impedance at 5Hz (Z5), and the area of ​​reactance (AX).

[0065] The spirometry data and forced oscillation lung function data in the data set may be parameters without inhaling bronchodilators, or parameters after inhaling bronchodilators (such as 400 ug salbutamol).

[0066] The FOT test should be performed before a spirometry test that requires deep breathing or after a spirometry test after at least 5 minutes of rest.

[0067] The spirometer data were checked for data quality control by professional respiratory physicians or technicians, and the physician checked the flow-volume curve and lung function values. The quality control standards were as follows: no hesitation at the start of exhalation, the ascending branch of the flow-volume curve rose steeply in the expiratory phase, the exhalation was explosive, the PEF peak appeared quickly, and the extrapolated capacity was <5% FVC or 0.15L (the maximum value); the descending branch curve of the expiratory phase was smooth, there was no interruption in the exhalation process, no coughing, no teeth or tongue blocking the mouthpiece, no air leakage, no glottal closure that affected the test, etc.; the exhalation time was at least 6 seconds. If the exhalation time was <6 seconds, the time-volume curve must show that the expiratory phase platform appeared and exceeded 1 second; inspiration: the inspiratory curve was full. Repeatability standards for multiple tests: the subjects were required to measure at least 3 times during the measurement process; the difference between the best and second-best values ​​of FEV1 and FVC was less than 0.15L.

[0068] Specifically, the FOT data is checked for data quality control by professional respiratory physicians, who check the volume-impedance curve, flow and pressure curves, and FOT values. The quality control standards are as follows: at least 3 respiratory cycles, stable tidal volume and respiratory rate during the acquisition period, no pauses or offsets in the volume curve, no sudden changes in the impedance curve or spike artifacts. Repeatability requirements: The repeatability requirements for 3 acceptable tests are: The coefficient of variation (CoV) of the resistance Rrs (such as R5) at the lowest oscillation frequency should be ≤10% for adults.

[0069] To establish a data set, it is necessary to select lung function data that have completed both spirometry and FOT examinations on the same day, with complete subject information and qualified quality control. If the same subject is found to have undergone multiple lung function tests over a period of time during retrospective data collection, the lung function data of the first test that can be traced back will be used.

[0070] In step 2, the data set of the development cohort is divided into a training set and a test set, the training set is used to build the model, and the test set is used to verify the discrimination and calibration of the model. Optionally, in step S2, the data set can be randomly split into a training set and a test set in a ratio of 1:1 or 7:3, or the data set can be split into independent training sets and test sets in chronological order.

[0071] In step 3, a multivariate linear regression was performed on the data set, and the outcome variable was the parameter used in the spirometry examination to determine airflow obstruction (diagnosis of COPD), that is, the measured value of FEV1 / FVC. The predictor variables were basic demographic parameters: gender, age, height, weight, and FOT parameters: R5, R20, Fres, X5, Z5, and AX. Among them, Z5 is the total respiratory impedance, which is composed of the vector sum of resistance (R5) and reactance (X5), that is, Therefore, Z5 may be collinear with R5 and X5 and can be removed before multi-factor regression, or it can be retained if it contributes greatly to the model according to the regression results.

[0072] The number of subjects required for modeling must be >500, because there may be multicollinearity between FOT parameters. If the modeling sample size is insufficient, it may lead to overfitting of the model. In order to avoid overfitting of the model, it is recommended to use machine learning algorithms including stepwise regression / optimal subsets / LASSO / ridge regression / elastic network method for modeling, and select the method with the best model performance in the test set to establish the final model.

[0073] Stepwise regression / optimal subset / LASSO / ridge regression / elastic network methods were used to screen variables with strong correlation with FEV1 / FVC, and linear regression was used for model fitting. The FEV1 / FVC prediction model based on FOT parameters and basic demographic data was constructed and the calculation formula was generated:

[0074] FEV1 / FVC=constant term+β1×age+β2×sex+β3×height+β4×weight+β5×R5+β6×R20+β7×X5+β8×Z5+β9×Fres+β 10 ×AX

[0075] The diagnostic criteria for COPD in step 4 are recommended to be FEV1 / FVC < 0.7 after inhalation of bronchodilators. If there is no condition for bronchodilator test, FEV1 / FVC < 0.7 before inhalation of bronchodilators can be used. Taking the diagnosis of COPD as the gold standard, the model-estimated FEV1 / FVC prediction value is used to calculate the ROC curve area AUC value for screening COPD defined by lung function, and the sensitivity, specificity, Youden index, accuracy, positive predictive value, negative predictive value, positive likelihood ratio and negative likelihood ratio of the model for COPD screening at different cut-off values ​​of FEV1 / FVC prediction value are calculated to determine the optimal cut-off value of the model for screening COPD defined by lung function. Optionally, a single cut-off value mode can be used to calculate the Youden index at different cut-off values. Youden index = sensitivity + specificity - 1, and the cut-off value when the Youden index is the largest is defined as the optimal cut-off value. Optionally, according to the different performances of the model in different scenarios, a double-cutoff value model can be used, defining sensitivity>90% as the exclusion cutoff value, specificity>90% as the inclusion cutoff value, and the gray area between the inclusion cutoff value and the exclusion cutoff value. Optionally, according to the different performances of the model in different scenarios, a double-cutoff value model can be used, defining negative predictive value (NPV)>90% as the exclusion cutoff value, positive predictive value (PPV)>90% as the inclusion cutoff value, and the gray area between the inclusion cutoff value and the exclusion cutoff value.

[0076] Optionally, according to the different performances of the model in different scenarios, a dual-cutoff value model can be used, taking the cutoff value when the Youden index is the maximum as the optimal cutoff value, to identify the best cutoff value of COPD (defined as FEV1 / FVC < 0.7 after inhalation of bronchodilator) as the inclusion cutoff value of the model, and to identify the best cutoff value of airflow obstruction (defined as FEV1 / FVC < 0.7 before inhalation of bronchodilator) as the exclusion cutoff value of the model. In order to include as many COPD patients as possible and exclude as many patients without airflow obstruction as possible.

[0077] If the double-cutoff value model is used, the predicted FEV1 / FVC value calculated by the prediction model can be used to determine whether the subject should undergo a spirometry test. If the predicted FEV1 / FVC value falls within the inclusion cutoff value, the subject is judged to have COPD; if the predicted value falls within the exclusion cutoff value, the subject is judged to be a healthy person; if the predicted value falls within the gray area, the subject is judged to be suspected of COPD and needs to undergo a spirometry test to confirm the diagnosis.

[0078] In step 5, internal validation was performed in the model development cohort. The prediction model formula was used to calculate the FEV1 / FVC model prediction values ​​of the subjects in the training set and the validation set, and the mean absolute error (MAE) and root mean square error (RMSE) between the model prediction values ​​and the actual measured values ​​were calculated. COPD defined by the diagnosis of lung function (FEV1 / FVC < 0.7 before or after the use of bronchodilators) was used as the predicted outcome. The ROC curve of the model in the training set and the validation set was drawn, and the area under the curve (AUC) value was calculated. Furthermore, the calibration curve of the model was drawn to present the deviation between the model prediction value and the actual measured value in the training set and the validation set.

[0079] Step 6: External validation of the model in other centers. The constructed COPD prediction model was externally validated using an external data set, which mainly included the following two steps: according to the data set structure for constructing the model, the validation data was sorted to maintain the consistency of data fields and units; the mean absolute error (MAE) and root mean square error (RMSE) of the model in the external data set were calculated, and the area under the curve (AUC) of the model for screening COPD in the external data set was calculated using the diagnosis of COPD defined by lung function (FEV1 / FVC < 0.7 before or after bronchodilators) as the gold standard, and the sensitivity, specificity, accuracy, positive predictive value, negative predictive value, positive likelihood ratio, and negative likelihood ratio of the model were calculated.

[0080] If the double boundary value model is used, the proportion of patients in the inclusion interval, exclusion interval and gray zone should also be calculated. The proportion of people whose predicted values ​​fall in the gray zone should not be too large, preferably less than 40% of the total number of people screened.

[0081] Example 1

[0082] This embodiment retrospectively analyzes the pulmonary function test databases of five hospitals from January 2020 to December 2023, including Beijing Chaoyang Hospital Affiliated to Capital Medical University, Beijing Jingmei Group General Hospital, Tongji Hospital Affiliated to Tongji Medical College of Huazhong University of Science and Technology, Jiangsu Provincial People's Hospital, and First Hospital of Shanxi Medical University. The database was screened for pulmonary function data that completed both spirometry and pulse oscillation examination (FOT examination using rectangular pulse signals) on the same day, and the subject information was complete and the quality control was qualified. If the same patient undergoes multiple pulmonary function tests over a period of time, the pulmonary function data of the first examination that can be traced back are used. Subjects aged <20 years and non-Chinese nationality were further excluded. The data of Beijing Chaoyang Hospital Affiliated to Capital Medical University, Tongji Hospital Affiliated to Tongji Medical College of Huazhong University of Science and Technology, and Beijing Jingmei Group General Hospital were used as the development cohort, and the data sets of the First Hospital of Shanxi Medical University and Jiangsu Provincial People's Hospital were used as external validation cohorts. The data set of the development cohort was randomly split into a training set and a test set 7:3 using R software. The training set was used to build the model, and the test set was used to verify the accuracy of the model. The flowchart of data set creation is shown in Figure 2 A total of 10,579 subjects were included in the training set, 4,534 subjects were included in the test set, and 9,180 subjects were included in the external validation set.

[0083] The basic information of the subjects in the training set and validation set is shown in Table 1. The average FEV1 / FVC before inhalation of bronchodilators in the training set was 0.74±0.11, and the incidence of airflow obstruction was 25.2%; the average FEV1 / FVC before inhalation of bronchodilators in the test set was 0.74±0.11, and the incidence of airflow obstruction was 25.7%; the average FEV1 / FVC in the external validation set was 0.76±0.10, and the incidence of airflow obstruction was 20.3%.

[0084] Table 1 Basic demographic and clinical characteristics of the subjects

[0085]

[0086]

[0087] Note: Variables are expressed as mean (standard deviation) or number of cases (percentage)

[0088] Nine variables, including gender, age, height, weight, R5, R20, Fres, X5 and AX, were used as predictor variables, and FEV1 / FVC before inhalation of bronchodilators was used as the outcome variable. In the training set, stepwise regression, optimal subsets, LASSO, ridge regression, and elastic network were used to build models. In the test set, the mean absolute error (MAE) and root mean square error (RMSE) of the models built by these modeling methods were compared. The model performance of different methods is shown in Table 2. The MAE and RMSE of various modeling methods were not much different, so we chose to use the simpler stepwise regression method to build the model.

[0089] Table 2 Model performance using different modeling methods

[0090]

[0091] Note: The table shows the performance of the prediction model in the test set. a The percentage of predicted FEV1 / FVC values ​​that are within 10.0% of the actual FEV1 / FVC value.

[0092] The MAE and RMSE of multivariate stepwise linear regression were 0.061 and 0.083, respectively. The results of multivariate stepwise linear regression are shown in Table 3. The regression coefficients were used to generate the prediction formula of FEV1 / FVC before inhalation of bronchodilator based on pulse oscillation lung function: FEV1 / FVC = -0.00137 × age + 0.0364 × sex (categorical variable, male is 0, female is 1) -0.00304 × height (in cm) + 0.00138 × weight (in kg) + 0.0858 × R5 (in kPa·s·L) -1 )-0.107×R20(unit: kPa·s·L -1 )-0.00334×Fres(unit: Hz)+0.182×X5(unit: kPa·s·L -1 )–0.0254×AX(unit: kPa·L -1 )+1.303Table 3 Multifactor linear regression results for predicting FEV1 / FVC

[0093]

[0094] Note: All lung function parameters were measured before inhalation of bronchodilators

[0095] Taking FEV1 / FVC < 0.7 before inhalation of bronchodilator as the diagnostic standard for airflow obstruction, the ROC curve of the model in the training set and the test set is shown in Figure 3The AUC value of the COPD screening model based on pulse oscillation lung function that we constructed to identify airflow obstruction was 0.816 (95% confidence interval 0.806-0.825) in the training set and 0.822 (95% confidence interval 0.808-0.836) in the test set, which was not much different from the training set. The calibration curve of the model in the training set and the test set (correlation scatter plot of FEV1 / FVC predicted value and actual value) is shown in Figure 4 The FEV1 / FVC predicted by the model was significantly correlated with the actual FEV1 / FVC value (Pearson correlation coefficient: training set r = 0.660, test set r = 0.658).

[0096] The sensitivity, specificity, Youden index, accuracy, positive predictive value, negative predictive value, positive likelihood ratio and negative likelihood ratio of the FEV1 / FVC predicted by the model for identifying airflow obstruction at different cut-off values ​​were calculated, and the results are shown in Table 4. The Youden index reached the maximum when the predicted value of FEV1 / FVC predicted by the model was 0.73, which was the best cut-off value for identifying airflow obstruction in this model. The sensitivity, specificity, accuracy, +LR, -LR, PPV and NPV for identifying airflow obstruction were 72.6%, 76.6%, 75.5%, 3.10, 0.36, 51.8% and 89.0%, respectively.

[0097] Table 4 Screening performance of the model for airflow obstruction at different cut-off values

[0098]

[0099] Note: The table shows the performance of the prediction model in the test set. * The optimal cutoff value when the Youden index is maximum.

[0100] Some subjects in the model development cohort (training set + validation set) and external validation cohort had undergone bronchodilator tests due to clinical diagnosis and treatment needs (generally, airflow obstruction before bronchodilator administration). The basic information of these subjects was retrospectively collected and shown in Table 5. FEV1 / FVC < 0.7 after inhalation of bronchodilators was used as the diagnostic standard for COPD defined by lung function. The screening performance of the model for COPD defined by lung function was calculated in these patients who had undergone bronchodilator tests. The AUC value of the pulse oscillation-based model we constructed to identify COPD defined by lung function was 0.849 in the development cohort (95% confidence interval was 0.836-0.862, Figure 5The sensitivity, specificity, Youden index, accuracy, positive predictive value, negative predictive value, positive likelihood ratio and negative likelihood ratio of the model for identifying COPD at different cut-off values ​​are shown in Table 6. The Youden index reaches the maximum when the predicted FEV1 / FVC value predicted by the model is 0.70, which is the best cut-off value for the model to identify COPD defined by lung function, with a sensitivity and specificity of 67.0% and 85.2%, respectively.

[0101] Table 5 Basic demographic and clinical characteristics of subjects who underwent bronchodilator trials

[0102]

[0103]

[0104] Note: Variables are expressed as mean (standard deviation) or number of cases (percentage)

[0105] Table 6 Screening performance of the model for COPD at different cut-off values

[0106]

[0107] Note: The table shows the performance of the prediction model in the development cohort. * The optimal cutoff value when the Youden index is maximum.

[0108] In order to include as many COPD patients as possible and exclude as many people without airflow obstruction as possible, we adopted a dual-cutoff model to identify the best cutoff value of COPD defined by lung function as the inclusion cutoff value of the model, and to identify the best cutoff value of airflow obstruction as the exclusion cutoff value of the model. The gray area is between the inclusion cutoff value and the exclusion cutoff value. In this study, the inclusion cutoff value is the model-predicted FEV1 / FVC < 0.70, and the exclusion cutoff value is the model-predicted FEV1 / FVC > 0.73. The model-predicted FEV1 / FVC between 0.70 and 0.73 is the gray area. If the subject's prediction result falls into the gray area, further spirometry examination is required to confirm the diagnosis of COPD. See the flowchart of our prediction model for screening COPD in clinical practice. Figure 6 .

[0109] The model was externally validated using the external validation set data. The mean absolute error (MAE) of the model in the external validation cohort was 0.076, and the root mean square error (RMSE) was 0.099, which was similar to the model development cohort. The FEV1 / FVC predicted by the model was strongly correlated with the actual FEV1 / FVC, with a Pearson correlation coefficient of r = 0.493, P < 0.001 ( Figure 7 A). The area under the ROC curve (AUC) value for the model to identify airflow obstruction in the external validation cohort was 0.790 (95% confidence interval, 0.779-0.802, Figure 7 B), the area under the ROC curve (AUC) for identifying COPD defined by lung function was 0.800 (95% confidence interval 0.780-0.820, Figure 7 C), indicating that the prediction model has good model performance in external validation. Using the double-cutoff mode, the sensitivity and specificity of the model for screening COPD in the external validation set were 74.8% and 88.0%, respectively.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, rather than to limit the protection scope of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the essence and scope of the technical solution of the present invention.

Claims

1. A method for establishing a COPD screening model based on forced oscillation lung function, characterized in that: The establishment method comprises the following steps: Step 1: Collect the spirometry and FOT examination data of the subjects on the same day and establish a clinical data set; the clinical data includes spirometry parameters, FOT parameters and basic demographic information; Step 2: Divide the collected clinical data into a training set and a test set. The training set is used to build the FEV1 / FVC prediction model, and the test set is used to verify the discrimination and calibration of the model. Step 3: Using the spirometry parameter FEV1 / FVC as the outcome variable, and the FOT parameter and demographic data as the predictor variables, the training set was subjected to multivariate linear regression, the regression coefficient β value of each variable was calculated, and the FEV1 / FVC prediction model based on the FOT parameter and basic demographic parameters was constructed, and the prediction formula was generated; Step 4: Calculate the predicted FEV1 / FVC value according to the prediction formula, calculate the sensitivity and specificity of the FEV1 / FVC prediction model for identifying COPD at different cutoff values, and find the optimal cutoff value of the model; Step 5: Conduct internal validation of the FEV1 / FVC prediction model in the development cohort and evaluate the model's screening performance for COPD in the test set; Step 6: Conduct external validation of the FEV1 / FVC prediction model in other centers to evaluate the model's screening performance for COPD in the external validation set.

2. The method for establishing a COPD screening model based on forced oscillation lung function according to claim 1, characterized in that: The basic demographic information described in step 1 includes gender, age, height, and weight, and the FOT parameters include resistance at 5 Hz, resistance at 20 Hz, resonance frequency, reactance at 5 Hz, respiratory impedance at 5 Hz, and reactance area.

3. The method for establishing a COPD screening model based on forced oscillation lung function according to claim 2, characterized in that: In step 3, the calculation formula of the FEV1 / FVC prediction model is as follows: FEV1 / FVC=constant term+β1×age+β2×sex+β3×height+β4×weight+β5×R5+β6×R20+β7×X5+β8×Z5+β9×Fres+β 10 ×AX In the above, β1 represents the regression coefficient of age, β2 represents the regression coefficient of gender, β3 represents the regression coefficient of height, β4 represents the regression coefficient of weight, β5 represents the regression coefficient of R5, R5 represents the resistance at 5 Hz, β6 represents the regression coefficient of R20, R20 represents the resistance at 20 Hz, β7 represents the regression coefficient of X5, X5 represents the reactance at 5 Hz, β8 represents the regression coefficient of Z5, Z5 represents the respiratory impedance at 5 Hz, β9 represents the regression coefficient of Fres, Fres represents the resonance frequency, β 10 represents the regression coefficient of AX, where AX represents the reactance area.

4. The method for establishing a COPD screening model based on forced oscillation lung function according to claim 3, characterized in that: In step 4, the diagnostic criteria for COPD were FEV1 / FVC < 0.7 after inhaled bronchodilators or FEV1 / FVC < 0.7 before inhaled bronchodilators. The diagnosis of COPD was taken as the gold standard, and the area under the ROC curve (AUC) of the FEV1 / FVC predictive value estimated by the model was calculated for screening COPD defined by lung function. The sensitivity, specificity, Youden index, accuracy, positive predictive value, negative predictive value, positive likelihood ratio and negative likelihood ratio of the model for COPD screening were calculated at different cutoff values ​​of FEV1 / FVC predictive values ​​to determine the optimal classification value of the model for screening COPD defined by lung function.

5. The method for establishing a COPD screening model based on forced oscillation lung function according to claim 4, characterized in that: When the double-cutoff value model is used in step 4, the criteria for judging whether the subject should undergo spirometry examination using the predicted FEV1 / FVC value calculated by the FEV1 / FVC prediction model are: If the predicted value of FEV1 / FVC falls within the inclusion boundary, the subject is judged to have COPD; if the predicted value falls within the exclusion boundary, the subject is judged to be a healthy person; If the predicted value falls into the gray area, the subject is judged to be suspected of COPD and needs to undergo a spirometry examination to confirm the diagnosis.

6. The method for establishing a COPD screening model based on forced oscillation lung function according to claim 4, characterized in that: The specific steps for internal validation of the FEV1 / FVC prediction model development cohort in step 5 are as follows: Step 501: Calculate the predicted FEV1 / FVC values ​​of the subjects in the training set and the test set using the calculation formula of the FEV1 / FVC prediction model; Step 502: Calculate the mean absolute error and root mean square error between the predicted FEV1 / FVC value and the actual measured FEV1 / FVC value, take the COPD defined by the diagnosis of lung function as the prediction result, draw the ROC curve of the FEV1 / FVC prediction model in the training set and the test set, and calculate the area under the ROC curve AUC value; Step 503: Draw a calibration curve of the FEV1 / FVC prediction model, and present the deviation between the predicted FEV1 / FVC values ​​and the actual measured FEV1 / FVC values ​​in the training set and the test set.

7. The method for establishing a COPD screening model based on forced oscillation lung function according to claim 6, characterized in that: In step 6, an external data set is used to perform external validation on the constructed FEV1 / FVC prediction model, including the following steps: Step 601: sorting the external data set according to the data set structure for constructing the FEV1 / FVC prediction model; Step 602: Calculate the mean absolute error and root mean square error of the FEV1 / FVC prediction model in the external data set, then use FEV1 / FVC<0.7 as the diagnostic standard for COPD, calculate the area under the curve AUC value of the FEV1 / FVC prediction model for screening COPD in the external data set, and calculate the sensitivity, specificity, Youden index, positive predictive value, negative predictive value, positive likelihood ratio and negative likelihood ratio of the FEV1 / FVC prediction model.

8. A system for constructing a COPD screening model based on forced oscillation lung function, characterized in that: The system includes: A data collection module is used to collect the spirometry and FOT examination data of the subjects on the same day and establish a clinical data set; wherein the clinical data includes spirometry parameters, FOT parameters and basic demographic information; Data processing, used to randomly allocate clinical data into training and test sets; The model building module uses the spirometry parameters in the training set as the outcome variables, the FOT parameters and demographic data as the predictor variables, performs multivariate linear regression on the training set, calculates the regression coefficient β value of each predictor variable, builds the FEV1 / FVC prediction model based on the FOT parameters and basic demographic parameters, and generates the calculation formula; Model optimization module, used to calculate the predicted FEV1 / FVC value according to the FEV1 / FVC prediction model calculation formula, and use FEV1 / FVC < 0.7 as the diagnostic standard for COPD, calculate the sensitivity and specificity of the model for identifying COPD at different cut-off values, and find the optimal cut-off value of the model; The validation module is used to perform internal validation in the FEV1 / FVC prediction model development cohort using the test set, and is also used to perform external validation of the FEV1 / FVC prediction model using an external dataset.

9. A COPD screening model construction system based on forced oscillation lung function according to claim 8, characterized in that: The calculation formula of the FEV1 / FVC prediction model constructed in the model building module is as follows: FEV1 / FVC=constant term+β1×age+β2×sex+β3×height+β4×weight+β5×R5+β6×R20+β7×X5+β8×Z5+β9×Fres+β 10 ×AX In the above, β1 represents the regression coefficient of age, β2 represents the regression coefficient of gender, β3 represents the regression coefficient of height, β4 represents the regression coefficient of weight, β5 represents the regression coefficient of R5, R5 represents the resistance at 5 Hz, β6 represents the regression coefficient of R20, R20 represents the resistance at 20 Hz, β7 represents the regression coefficient of X5, X5 represents the reactance at 5 Hz, β8 represents the regression coefficient of Z5, Z5 represents the respiratory impedance at 5 Hz, β9 represents the regression coefficient of Fres, Fres represents the resonance frequency, β 10 represents the regression coefficient of AX, where AX represents the reactance area.

Images