CN115775630A - Postoperative lung complication probability prediction method based on sleep stage data before operation - Google Patents
Postoperative lung complication probability prediction method based on sleep stage data before operation Download PDFInfo
- Publication number
- CN115775630A CN115775630A CN202310096690.4A CN202310096690A CN115775630A CN 115775630 A CN115775630 A CN 115775630A CN 202310096690 A CN202310096690 A CN 202310096690A CN 115775630 A CN115775630 A CN 115775630A
- Authority
- CN
- China
- Prior art keywords
- characteristic data
- data set
- data
- characteristic
- physiological
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The disclosure belongs to the field of lung complication prediction, and particularly relates to a postoperative lung complication probability prediction method based on sleep stage data, which comprises the following steps: acquiring a pre-operation physiological clinical characteristic data set of a predicted person, wherein the pre-operation physiological clinical characteristic data set comprises a pre-operation physiological characteristic data set and a pre-operation clinical characteristic data set; the obtaining of the pre-operative clinical characteristic data set comprises collecting continuous physiological signals of the sleep stage of the predicted person based on the pre-operative clinical characteristic data setExtracting preoperative clinical characteristic data of the forecasted person according to the continuous physiological signals; predicting a function based on the set of physiological characteristic data and clinical characteristic data
Obtaining a predicted post-operative complication probability of the lung; wherein
Array formed for input characteristic data of predicted personxThe probability of a complication of (2),θa vector formed for the characteristic coefficients is formed,
is the characteristic coefficient corresponding to the n-th item of characteristic data,xa vector formed for the feature data. To conveniently and relatively accurately assess the probability of pulmonary postoperative complications prior to surgery.
Description
Technical Field
The disclosure belongs to the field of lung complication prediction, and particularly relates to a postoperative lung complication probability prediction method based on sleep stage data before operation.
Background
Post-operative pulmonary complications (PPCs) are associated with increased postoperative mortality, prolonged hospital stay and increased medical expenditure, and are one of the major causes of poor post-operative prognosis in surgical patients. The incidence of PPCs varies considerably among the different operative populations, while the cardiac surgical population is exposed to a higher risk of PPCs. Over 4000 million people worldwide suffer from mitral valve or aortic valve diseases, heart valve operation is one of the great risks in heart surgery, and extracorporeal circulation during the operation process is easy to cause systemic inflammatory response and oxidative stress reaction, so that lung ischemia-reperfusion injury is caused. The main reasons for the patients with valvular heart disease after operation to re-enter ICU are 40% of the pulmonary complications after operation and 23% of the respiratory failure after operation, which are common types of PPCs.
Before the operation, the risk probability of the PPCs is evaluated, so that important reference information of whether to perform the operation and when to perform the operation is provided for the predicted person. Therefore, how to effectively predict the risk of PPCs occurring in the heart valvulopathy operation group before the operation is performed, so that corresponding early warning is performed, clinical intervention measures are intervened as soon as possible, the poor outcome of the PPCs of the group is reduced, and the clinical problem to be solved is urgently needed. Cavayas et al evaluated preoperative diaphragm function in a cardiac surgical population via diaphragm ultrasound and found that a decreasing maximum diaphragm thickness fraction (< 38.1%) during inspiration helped identify the risk of developing PPCs in this population (OR =4.9; 95% CI, 1.81-13.50; p = 0.002), but this required not only extensive clinical ultrasound diagnostic experience, but also the provision of more expensive ultrasound diagnostic equipment. In addition, other researchers have developed many risk prediction models for identifying patients at high risk for PPCs, thereby achieving better perioperative management. The risk of delivering PPCs in the catalonia surgical patients was evaluated, the patients were classified into low, medium and high risk groups, and seven independent variables of the regression model included whether the preoperative peripheral blood oxygen saturation was less than 96%, whether there was respiratory infection in the previous month, age, preoperative anemia (< 100 g/dl), surgical site, surgical time (> 2 h), whether emergency surgery was performed, and people such as valentii n Mazo verified the external validity of the above model in the european population, showing better discriminatory ability. Another prospective multicenter cohort study focused exclusively on PPCs risk stratification in epigastric incisional patients, who identified five independent risk factors in their regression model, including anesthesia time, surgical category, respiratory complications, current smoking and predicted maximum oxygen uptake, with scores below 2.02 associated with high risk of PPC [ OR (CI) =8.41 (3.33-21.26) ], but this model still required external validation.
The existing risk scoring methods all depend on indexes such as previous diagnosis (smoking history, COPD (chronic obstructive pulmonary disease) history, preoperative sepsis, existing chest cavity or upper abdominal operation incision and the like), clinical examination results (pneumonia), examination results (preoperative albumin, hemoglobin, blood urea nitrogen level and the like), and the discrimination of the risk scoring models in the heart valve operation population is still questioned. Therefore, a means with strong operability, low cost and good prediction effect is clinically needed for predicting the risk of the PPCs of the patient with the valvular heart disease before the operation.
Disclosure of Invention
The present disclosure is made based on the above-mentioned needs of the prior art, and the technical problem to be solved by the present disclosure is to provide a method for predicting postoperative pulmonary complications probability based on sleep stage data before operation to conveniently and relatively accurately evaluate the probability of pulmonary postoperative complications before operation, so as to provide a reference for a predicted person.
In order to solve the above problem, the technical solution provided by the present disclosure includes:
the invention provides a postoperative lung complication probability prediction method based on sleep stage data before operation, which comprises the following steps: acquiring a pre-operation physiological clinical characteristic data set of a predicted person, wherein the pre-operation physiological clinical characteristic data set comprises a pre-operation physiological characteristic data set and a pre-operation clinical characteristic data set; acquiring the pre-operative clinical characteristic data set comprises acquiring continuous physiological signals of the sleep stage of the predicted person, wherein the continuous physiological signals comprise: continuous single lead electrocardiosignals, continuous thoracoabdominal respiration signals and continuous sleep state signals; extracting pre-operative clinical feature data of the predicted person based on the continuous physiological signals, wherein the pre-operative clinical feature data comprises a first feature data set of heart rate variability obtained at least based on NN intervals obtained by calculation of the continuous single-lead electrocardiosignals, a second feature data set obtained at least based on the continuous thoracoabdominal respiration signals and a third feature data set obtained at least based on continuous sleep state signals; based on at least one feature data of each of the physiological feature data set, the first feature data set, the second feature data set and the third feature data set, predicting a function
Obtaining a predicted post-operative complication probability of the lung; wherein, the first and the second end of the pipe are connected with each other,
，
，
array formed for input predicted person characteristic dataxThe probability of a complication of (2),θa vector formed for the characteristic coefficients is formed,
is a constant number of times, and is,
is the characteristic coefficient corresponding to the n-th item of characteristic data,xa vector formed for the feature data,
，
is the nth characteristic data.
The relation between the characteristic data and the postoperative complications of the lung is discovered in the method, the probability of postoperative complications can be accurately obtained by acquiring the required data and establishing a good prediction model, so that whether the operation is performed or not is guided, and which aspect of adjustment is required to perform the operation, the success rate of the operation is ensured, and the risk is avoided to a great extent.
Preferably, the feature data in the first feature data group are: the number of interval between two NN intervals before and after being more than 50ms, the average value of the whole NN interval, high-frequency energy, the ratio of low frequency to high frequency and arrhythmia load.
Preferably, the feature data in the second feature data group are: mean minute ventilation during sleep, mean breathing rate during sleep, and mean inspiratory time during sleep.
Preferably, the feature data in the third feature data group calculated based on the sleep state signal is: the REM sleep duration of the rapid eye movement period accounts for the percentage of the whole sleep duration, the deep sleep duration accounts for the ratio, and the effective blood oxygen duration accounts for the ratio.
Preferably, the characteristic data in the preoperative physiological characteristic data set is: surgical procedure data, age data, and preoperative pulmonary artery diameter data.
Preferably, the prediction function is determined by: constructing a first predictive model from a feature data set associated with a pulmonary post-operative complication, the feature data set having an array of m feature data; obtaining an optimal characteristic coefficient of the first prediction model through iteration; arranging the optimal characteristic coefficients of the first prediction model according to the size sequence, and deleting the minimum characteristic coefficient and the characteristic data item corresponding to the minimum characteristic coefficient; circularly carrying out the steps of construction, iteration, arrangement and deletion of a re-prediction model based on the residual characteristic data until the complication prediction value of the Nth prediction model is reduced by a first threshold value after a certain characteristic data is deleted; and reserving the m-N items of feature data and the feature coefficients corresponding to the reserved feature data as final feature data and final feature coefficients respectively to obtain the prediction function.
Preferably, the prediction model comprises a logistic regression model, and the likelihood function of the dataset based on the logistic regression model is represented as:
wherein m is the number of the continuous physiological clinical parameter arrays in the data set,
is the ith continuous physiological clinical parameter array,
is composed of
The label which is corresponding to the label is provided,Yis 0 or 1.
Preferably, the cost function obtained based on the likelihood function is expressed as:
wherein m is the number of the continuous physiological clinical parameter arrays in the data set,
is the ith continuous physiological clinical parameter array,
is composed of
The label which is corresponding to the label is provided,Yis 0 or 1.
Preferably, the characteristic coefficients are initialized through gradient descent and are gradually updated until the characteristic coefficients optimal for the characteristic data in the continuous array of physiological clinical parameters are obtained
Expressed as:
wherein, in the process,
is as followsjThe characteristic coefficient corresponding to the item characteristic data, alpha is the learning rate,
as a cost function.
Preferably, the first threshold value includes 10% or more.
Preferably, the characteristic data extracted from the respiratory signal is obtained by performing smoothing filtering on the acquired respiratory signal, removing an abnormal value, and detecting a peak and a trough.
Preferably, the NN interval derived heart rate variability feature data is derived from the acquired electrocardiogram signal and the position of the R-wave peak is detected.
Preferably, the method for forming the prediction function further includes: validating the predictive model, including: averagely dividing the data set into 5 parts, namely D1, D2, D3, D4 and D5, taking D1-D4 as a verification set and D5 as a verification set, and calculating a first-folding performance parameter; sequentially circulating for 4 times, and calculating performance parameters from the second folding to the fifth folding; a degree of accuracy of the prediction function is calculated based on the performance parameter to determine whether the prediction function is appropriate.
Compared with the prior art, the method and the device determine the probability of the sleep stage physiological characteristic data and the pulmonary postoperative complications through theoretical analysis, empirical summary and model calculation, can obtain a high-accuracy probability result of the pulmonary postoperative complications through data testing in the sleep stage, and are convenient for a predicted person to obtain accurate evaluation. In addition, the probability prediction function of the pulmonary postoperative complications is built based on the four characteristic data groups, compared with the existing model which is obtained based on theoretical analysis and data checking, the probability prediction function is comprehensive in consideration factors, the situation that the data are not accurate and simultaneously selected due to external environment or psychological factors is avoided, the characteristic correlation strong data size is appropriate, and the accuracy and the calculation speed of the model are improved.
Drawings
In order to more clearly illustrate the embodiments of the present specification or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the embodiments of the present specification, and other drawings can be obtained by those skilled in the art according to the drawings.
FIG. 1 is a flow chart of steps in a method of predicting post-operative complications of the lung based on sleep phase data in an embodiment of the present disclosure;
FIG. 2 is a flow chart illustrating steps of a method for forming a predictive model in an embodiment of the present disclosure.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
In the description of the embodiments of the present disclosure, it should be noted that, unless otherwise explicitly stated or limited, the term "connected" should be interpreted broadly, and may be, for example, a fixed connection, a detachable connection, or an integral connection, which may be a mechanical connection, an electrical connection, which may be a direct connection, or an indirect connection via an intermediate medium. The specific meaning of the above terms in the present disclosure can be understood as a specific case by a person of ordinary skill in the art.
The terms "top," "bottom," "above … …," "below," and "above … …" as used throughout the description are relative positions with respect to components of the device, such as the relative positions of the top and bottom substrates inside the device. It will be appreciated that the devices are multifunctional, regardless of their orientation in space.
For the purpose of facilitating understanding of the embodiments of the present application, the following detailed description will be given with reference to the accompanying drawings, which are not intended to limit the embodiments of the present application.
The present embodiment provides a method for predicting postoperative complications of the lungs based on sleep phase data, as shown in fig. 1-2.
Acquiring a pre-operation physiological clinical characteristic data set of a predicted person, wherein the pre-operation physiological clinical characteristic data set comprises a pre-operation physiological characteristic data set and a pre-operation clinical characteristic data set.
The pre-operation physiological clinical characteristic data of the predicted person is considered to be an important physiological index capable of reflecting the cardiopulmonary function of the predicted person, and the PPCs are closely related to the cardiopulmonary function, so that the influence of the parameters of the body organ of the predicted person on the PPCs occurrence probability can be considered by selecting the pre-operation physiological characteristic data set of the predicted person as the influence factor for evaluating the risk probability of the PPCs. In the present embodiment, the order of the steps for acquiring the pre-operation physiological characteristic data set of the predicted person may be flexibly set, and the acquisition of the pre-operation physiological characteristic data set of the predicted person may be performed only in the first step as long as the data is acquired when the data is needed.
In this embodiment, the following characteristic data may be included: surgical procedure (operation), age (age), cardiac functional classification criteria (NYHA), european cardiovascular surgical risk factor Score (Euro Score), left ventricular inside diameter, left ventricular enlargement, left atrial inside diameter, left atrial enlargement, right ventricular inside diameter, right ventricular enlargement, right atrial inside diameter, right atrial enlargement, left ventricular end diastolic volume, left ventricular end diastolic structural change parameter, left ventricular end systolic volume, left ventricular end systolic structural change parameter, left ventricular ejection fraction, left ventricular systolic functional change parameter, pulmonary artery diameter (pulmonary).
These data are theoretically considered to have a great influence on the occurrence of PPCs, but those data are strongly related to PPCs and those data are weakly related to PPCs, and it is not explicitly disclosed in the existing research, on one hand, because these data are theoretically closely related to the cardiopulmonary function, the correlation of a specific data cannot be easily rejected, and on the other hand, because the existing PPCs determination method is often limited to a certain angle and cannot comprehensively evaluate the factors that affect PPCs in whole, the correspondence between these parameters and PPCs is not completely clear, so that the degree of influence of these parameters on PPCs cannot be accurately evaluated.
In this embodiment, it is preferable that the data can represent parameters strongly correlated with the risk of PPCs occurrence in the physiological characteristic data set of the predicted subject, and the data can be selected as the representative data of the physiological characteristic data set to significantly reduce the calculation amount of PPCs prediction, improve the prediction speed, and ensure the accuracy of prediction.
This benefits primarily from the overall and rational predictive model considerations of the present application and the accuracy and scientificity of screening feature data in feature data sets. This will be developed in the following description of the present embodiment.
In addition to the pre-operative physiological characteristics of the predicted person, the apnea condition of the predicted person during sleep is considered to have strong correlation with the PPCs.
Sleep apnea syndrome (OSA) is a clinical syndrome in which a series of pathophysiological changes occur in the body due to hypoxemia, hypercapnia, and sleep interruption caused by repeated apneas or hypopneas in a sleep state mainly caused by obstruction. OSA patients have a significantly increased risk of perioperative postoperative complications, higher risk of postoperative respiratory failure, cardiac complications, hypoxemia, prolonged hospital stays, and more frequent ICU metastases. OSA patients are at increased risk of respiratory depression due to their higher susceptibility to sedatives and opioids, and furthermore, after major surgery, abnormalities in central control of breathing may lead to a reduction in the patient's ventilatory response to hypoxia and hypercapnia, further leading to more severe respiratory complications in OSA patients. The probability of complex arrhythmia of patients with severe OSA is 2 to 4 times higher than that of patients without OSA, and moreover, preoperative low blood oxygen level of OSA patients also reflects the heart and lung function status, and is a powerful predictor of postoperative pulmonary complications.
The physiological parameters during sleeping have high correlation with the risk of PPCs, if the physiological parameters of the human body during sleeping are measured, the heart and lung function parameters of the evaluated person in the sleeping stage are detected, and the method has strong reference significance for judging the risk probability of PPCs of the evaluated person.
Therefore, in this embodiment, the method for predicting probability of postoperative pulmonary complications based on sleep stage data includes: acquiring a continuous physiological signal of a sleep stage of a predicted person, the continuous physiological signal comprising: continuous single lead electrocardio signals, continuous thoracoabdominal respiration signals and continuous sleep state signals.
In this embodiment, the single lead ECG signal can be implemented by using a portable ECG detecting device, such as a portable ECG detecting device clamped or attached to a predetermined position of the body, to detect the original single lead ECG signal. The thoracoabdominal respiration signal can be realized by using a belt type respiration detection sensor in the prior art. The detection of the sleep state can be acquired by a portable sensor for detecting eye movement and a portable blood oxygen detection sensor matched with a breath detection sensor.
The signal acquisition devices acquire continuous related physiological signals, preferably continuous signals of a predicted person at night, and the continuous signals can reflect physiological characteristics for a longer time and change of the physiological characteristics during detection, so that the continuous signals can be used for better evaluating physiological function indexes related to the PPCs.
However, since the signal detection is only a general detection signal and not only detection for the risk of PPCs, it is necessary to process these data in order to more accurately evaluate the risk probability of PPCs occurrence, and to obtain feature data having a stronger correlation with PPCs, that is, pre-operation clinical feature data of a predicted person based on these continuous signal processing analysis, thereby improving the accuracy and speed of the prediction of PPCs.
In this embodiment, based on theoretical studies and summary of experimental data, the continuous physiological signals acquired in the previous step are analyzed and processed to obtain three pre-operation clinical characteristic data sets of the predicted person, namely, a first characteristic data set of heart rate variability obtained based on NN intervals calculated by the continuous single lead electrocardiogram signals, a second characteristic data set calculated based on the continuous thoracoabdominal respiration signals, and a third characteristic data set calculated based on the continuous sleep state signals.
Wherein the NN interval is an interval between normal R peaks in an electrocardiogram. The NN interval derived heart rate variability feature dataset may comprise: the number of the preceding and following NN interval is more than 50ms (nni _ 50), the number of the preceding and following NN interval is more than 20ms (nni _ 20), the average value of the whole NN interval (mean _ nni), the median value of the whole NN interval (mean _ nni), the standard deviation of the whole NN interval (nni _ std), the root mean square of the difference between two adjacent NN intervals (rmssd), the ultra-low frequency energy (vlf), the low frequency energy (lf), the high frequency energy (hf), the ratio of the low frequency to the high frequency (lf _ hf _ ratio), the difference between the maximum NN interval and the minimum NN interval in the sleep stage (min _ max _ nni), and the arrhythmia load (ar _ burden) are calculated by the number of the difference between the preceding and following NN intervals being more than 160ms, and the whole NN interval accounts for the percentage of the number of the whole NN intervals.
The characteristic data can be obtained by acquiring an original single-lead ECG (electrocardiogram) signal obtained in the step of acquiring and detecting the position of an R wave peak, specifically, the position of the R wave peak is detected by using a Hamilton method for the original ECG signal, wherein the position of the detected R wave peak is shown by a drawing point, an NN interval sequence of a whole night is obtained after the R wave detection is carried out on the ECG signal of the whole night, and the characteristic data is obtained from the RR interval sequence.
A second feature data set calculated based on the continuous thoracoabdominal respiratory signal comprises: mean respiratory rate (br _ mean), standard deviation respiratory rate (br _ std), mean inspiratory time (TI _ mean), standard deviation inspiratory time (TI _ std), mean inspiratory time as a percentage of the total breath duration (TI _ ratio _ mean), mean minute ventilation (min _ ven _ in _ mean). And calculating the power spectral density by adopting an FFT method after the NN interval signal difference value is 2Hz, wherein: the value of the high-frequency component (hf) is between 0.15 and 0.4 Hz; the value of the low-frequency component (lf) is between 0.04 and 0.15Hz; the total power (tf) is between 0.04 and 0.4 Hz; the value of the medium frequency component (hf) is between 0.1 and 0.15Hz; the value of the T-low frequency component (tlf) is between 0.04 and 0.1Hz, and the value of the extremely-low frequency component (vlf) is between 0.0033 and 0.04Hz.
The second characteristic data set can be obtained by performing smooth filtering processing on the acquired original thoracoabdominal respiration signal all night during sleep and detecting a peak and a trough after removing an abnormal value.
The third feature data set calculated based on the sleep state signal includes: sleep duration (sleep _ len), percentage of REM sleep duration in fast eye movement period to total sleep duration (REM _ per), deep sleep duration ratio (deep _ per), light sleep duration ratio (light _ per), wake times in sleep (wake _ times), effective blood oxygen duration ratio (used _ per), sleep apnea hypopnea index (ahi), duration ratio of blood oxygen below 90% during sleep (spo 2_ 90), duration ratio of blood oxygen below 85% during sleep (spo 2_ 85), minimum blood oxygen value during sleep (spo 2_ min)
The third characteristic data calculated based on the sleep state signal may be calculated based on the third sleep data by using a sleep result evaluation method in the prior art, or may be directly obtained from an existing sleep evaluation report if the evaluation report is generated and the data is included.
In summary, the physiological characteristic data set related to sleep has 47 characteristic data, that is: the number of preceding and following NN interval greater than 50ms (nni _ 50), the number of preceding and following NN interval greater than 20ms (nni _ 20), the average of the whole NN interval (mean _ nni), the median of the whole NN interval (mean _ nni), the standard deviation of the whole NN interval (nni _ std), the root mean square of the difference between two adjacent NN intervals (rmssd), the ultra low frequency energy (vlf), the low frequency energy (lf), the high frequency energy (hf), the ratio of low and high frequency (lf _ hf _ ratio), the difference between the maximum NN interval and the minimum NN interval of the sleep phase (min _ max _ nni), the arrhythmia load (ar _ burden) are calculated by the number of preceding and following NN interval difference greater than 160ms, the percentage of the whole NN interval mean, the average of the breathing frequency (br _ mean _ n), the breathing frequency difference (mean _ nni), the arrhythmia load (ar _ burden) and the total inspiration Time (TI). And calculating the power spectral density by adopting an FFT method after the NN interval signal difference value is 2Hz, wherein: the value of the high-frequency component (hf) is between 0.15 and 0.4 Hz; the value of the low-frequency component (lf) is between 0.04 and 0.15Hz; the total power (tf) is between 0.04 and 0.4 Hz; the value of the medium frequency component (hf) is between 0.1 and 0.15Hz; the value of the T-low frequency component (tlf) is between 0.04 and 0.1Hz, the value of the very low frequency component (vlf) is between 0.0033 and 0.04Hz, the sleep time (sleep _ len), the percentage of the REM sleep time in the rapid eye movement period to the whole sleep time (REM _ per), the proportion of the deep sleep time (deep _ per), the proportion of the light sleep time (light _ per), the wake-up times in the sleep process (wake _ times), the proportion of the effective blood oxygen time (used _ per), the sleep apnea low ventilation index (ahi), the proportion of the blood oxygen in the sleep period to the blood oxygen of less than 90% (spo 2_ 90), the proportion of the blood oxygen in the sleep period to the blood oxygen of less than 85% (spo 2_ 85), and the lowest blood oxygen value in the sleep period (spo 2_ min).
Although all of the 47 characteristic data may be related to the PPCs theoretically, the speed of calculation is affected by taking as input so much data, and accurate evaluation is required to determine which data in the 47 characteristic data is strongly related to the PPCs and which data is not strongly related to the risk probability of the PPCs occurring. Therefore, in the present embodiment, for the purpose of screening the strongly correlated data among the 47 items of data and increasing the calculation speed of the prediction model, the prediction model is formed by the following method.
Acquiring a data set, wherein the data set comprises a plurality of continuous physiological clinical parameter arrays of sleep stages associated with pulmonary postoperative complications, and the continuous physiological clinical parameter arrays comprise acquired physiological characteristic data and clinical characteristic data extracted based on analysis of single lead electrocardiosignals, thoracoabdominal respiration signals and blood oxygen signals; the physiological clinical parameter array is composed of a plurality of characteristic data, wherein the characteristic data are heart rate variability characteristic data obtained through NN intervals, characteristic data extracted from respiratory signals, characteristic data obtained from sleep analysis results and preoperative physiological characteristic data.
Further, the data set used in this embodiment includes multidimensional analysis of sleep conditions, electrocardio, respiration, and blood oxygen during the whole sleep period, features are extracted, and the features are screened to finally obtain the data set. The data set comprises a plurality of continuous physiological clinical parameter arrays of sleep stages and associated with pulmonary postoperative complications, and a plurality of characteristic data included in the physiological clinical parameter arrays comprise four types, namely: the NN interval obtains heart rate variability feature data; characteristic data extracted from the thoracoabdominal respiratory signals; and feature data and preoperative physiological feature data extracted from the sleep analysis result. A plurality of four classes
And establishing a prediction model based on the data set to obtain final feature data forming the prediction function and a final feature coefficient corresponding to the final feature data.
The establishing of the prediction model based on the data set comprises the following steps:
a first predictive model is constructed from a feature data set associated with a pulmonary postoperative complication, the feature data set having an array of M feature data.
The outcome based on the complication requires that the predictive model eventually output a number between 0~1 and is determined to be 1 if the function value is greater than 0.5 and 0 otherwise. Meanwhile, a parameter to be determined is needed in the function, and the parameter can be accurately predicted for data in a training set by utilizing sample training. Thus, the prediction model is set as a logistic regression model, and the sigmoid function is set as a prediction function, and is expressed as:
wherein, the first and the second end of the pipe are connected with each other,
array formed for input characteristic dataXThe probability of a complication of (2) is,Θa vector formed for the coefficients of the feature,
the characteristic coefficient corresponding to the N-th item of characteristic data,Xa vector formed for the feature data,
，
is the Nth characteristic data.
In this embodiment, the data set includes 47 items of feature data, i.e., N =47, and the feature coefficients are initially assigned randomly.
The likelihood function of the dataset based on a logistic regression model is represented as:
wherein m is the number of continuous physiological clinical parameter arrays in the data set,
is the ith continuous physiological clinical parameter array,
is composed of
The corresponding label is marked with a corresponding label,Yis 0 or 1.
The cost function based on the likelihood function is expressed as:
initializing the characteristic coefficients by gradient descent and gradually updating until the characteristic coefficients optimal for the characteristic data in the continuous physiological clinical parameter array are obtained, which is expressed as:
wherein the content of the first and second substances,
is as followsjThe feature coefficient corresponding to the item feature data, alpha is the learning rate,
as a cost function.
Taking the above embodiment as an example, by means of gradient descent, iteration is continuously performed to approximate the feature coefficient that minimizes the cost function, that is, the optimal feature coefficient under the first prediction model.
After the characteristic coefficients are obtained through the optimization process, the characteristic coefficients are arranged according to the size sequence, and the minimum characteristic coefficient and the characteristic data corresponding to the minimum characteristic coefficient are deleted. Specifically, the obtained optimal feature coefficients under the first prediction model are arranged from large to small, and feature data corresponding to the 47 th feature coefficient are removed, that is, feature data with the smallest weight relative to postoperative pulmonary complications are removed, that is, the feature data have the smallest influence on postoperative complications.
And reconstructing the second prediction model by the data set with the characteristic data deleted. Specifically, the above process is repeated with the dataset from which one feature data is eliminated as a new dataset, that is, the second prediction model is newly formed based on the dataset having 46 feature data.
The prediction function of the second prediction model is a Sigmoid function, the construction process of the second prediction model is the same as that of the first prediction model, and the difference between the two prediction models is that the feature data deleted after the first iterative ordering is not carried out in the data set of the second prediction model, so that the finally obtained optimal feature coefficient is different from that obtained by the first prediction model. And then, sequencing the optimal characteristic coefficients according to the size sequence, and removing the characteristic data corresponding to the minimum characteristic coefficient from the data set. The elimination process is also used for deleting the factors which have the minimum influence on the capability of postoperative pulmonary complications so as to discharge data which does not need to be acquired, so that the acquisition process is simplified, the operation is more convenient and quicker, and meanwhile, the data volume can be effectively reduced.
And circularly reconstructing an Nth' prediction model based on the residual characteristic data and arranging and removing the characteristic data until the complication prediction value of the Qth prediction model is reduced by a first threshold value after a certain characteristic data is deleted. The first threshold comprises 10% or greater.
And reserving the M-Q item feature data and the feature coefficient corresponding to the reserved feature data as final feature data and a final feature coefficient respectively to obtain the prediction function.
Compared with models such as a neural network and the like, the prediction result is obtained more quickly and intuitively through the logistic regression model on the premise of ensuring the probability of the predicted complications.
Based on the above embodiment of the present embodiment, the input of the finally obtained prediction model is 15 items of feature data, which are: nni _50, mean _ nni, lf _ hf _ ratio, hf, ar _ garden, min _ ven _ in _ mean, br _ mean, TI _ mean, rem _ per, deep _ per, used _ per, ahi, operation, age, and pulmonary.
Namely, the characteristic data in the heart rate variability characteristic data group obtained by the NN interval are: the number of the interval between the two NN intervals before and after the interval is more than 50ms (nni _ 50), high frequency energy (hf), ratio of low frequency to high frequency (lf _ hf _ ratio), and arrhythmia load (ar _ burden).
The characteristic data in the second characteristic data group calculated based on the continuous thoracoabdominal respiration signals are: the minute ventilation average (min _ ven _ in _ mean), expressed as the average of the breathing frequency during sleep (br _ mean), and expressed as the average of the breathing frequency during sleep (TI _ mean).
The feature data in the third feature data group calculated based on the sleep state signal is: the REM sleep duration of the fast eye movement period is a percentage of the entire sleep duration (REM _ per), the deep sleep duration ratio (deep _ per), and the effective blood oxygen duration ratio (used _ per).
The characteristic data in the physiological characteristic data set are as follows: procedure (operation), age (age), and pulmonary artery diameter (pulmonary).
Obtaining a prediction model according to the finally determined 15 items of feature data and a final feature coefficient corresponding to the final feature data, wherein a prediction function of the prediction model is represented as:
wherein, the first and the second end of the pipe are connected with each other,
array formed for input characteristic dataxThe probability of a complication of (2),θa vector formed for the characteristic coefficients is formed,
is the characteristic coefficient corresponding to the n-th item of characteristic data,xa vector formed for the feature data,
，
is the nth characteristic data.
In the above embodiment of the present embodiment, the prediction function of the prediction model is represented as: PPC =0.0867386 × operation +0.0024807 × ahi +0.0145861 × used _ per-0.0569415 × rem _ per-0.000001 × nni _50-0.000006 × min _ ven _ in _ mean +0.0218722 × deep _ per +0.0027144 × mean _ nni-0.0950354 × br _ mean-0.0469357 × 1f \ hf \ ratio of microwave oven 0.0127289 × min _ in _ cv +0.0181016 × TI _ mean +0.00003 × hf-0.0601069 × 5252525272 × render 7972 × 7945 zonstar.
After the 15 characteristics pass through the logistic regression model, the output result is a percentage value, and the representative meaning is as follows: the incidence of postoperative pulmonary complications was judged by Melbourne score.
In order to obtain more reliable and stable model evaluation, a training set and a verification set are generated in a 5-fold cross-validation mode. The method specifically comprises the following steps:
the entire data set was divided into 5 equal parts, D1, D2, D3, D4, D5.
And taking D1 to D4 as a training set and D5 as a verification set, and calculating a first-folding performance parameter.
And sequentially circulating for four times, and calculating the model performance evaluation of two to five folds.
And calculating the accuracy degree of the prediction function based on the performance parameters, and evaluating the performance of the prediction model.
Furthermore, the performance parameters obtained by the 5-fold cross validation method evaluate the prediction capability of the model by calculating the Area (AUC) under the ROC curve (referring to the receiver operation characteristic curve), and in addition, the evaluation indexes of the Accuracy (ACC), the F1 score (F1), the Precision (Precision) and the Recall (Recall) are also calculated, so that the performance of the model is evaluated in multiple aspects.
Specifically, the following confusion matrix is constructed according to the result output by the prediction model and the real label condition:
ROC is a tool for measuring the imbalance in classification, and the ROC curve and AUC are often used to evaluate the merits of a binary classifier. The abscissa represents the FPR (False positive rate) and the ordinate represents the True positive rate TPR (True positive rate). The AUC (Area Under Curve) is defined as the Area Under the ROC Curve, and since the ROC Curve is generally located above the line y = x, the value range is between 0.5 and 1, and the AUC is used as an evaluation index because the ROC Curve cannot clearly indicate which classifier has a better effect in many cases, and the larger the AUC is used as a numerical value, the better the classifier has a better effect. The abscissa and ordinate are respectively expressed as:
the accuracy ACC is expressed as:
the F1 score is expressed as:
through the verification of the performance parameters, the prediction model is evaluated, and the result is expressed as:
the verification result shows that the prediction model finally obtained in the embodiment has good classification performance, and the accuracy and stability of the prediction model are shown through the obtained other data. F1 is a harmonic mean value of the precision rate and the recall rate to measure the overall performance of the classifier, and based on the data, the overall performance of the prediction model in the embodiment is good.
After 15 data before and after the operation of the forecasted person are obtained, the probability of postoperative lung complications of the forecasted person can be obtained by combining the expression of the PPC, the probability of postoperative complications of the forecasted person can be accurately estimated through the method, so that whether the operation is performed or not is guided, and the operation can be performed only by adjusting which aspect is required, the success rate of the operation is guaranteed, and the risk is avoided to a great extent.
The above-mentioned embodiments, objects, technical solutions and advantages of the present application are described in further detail, it should be understood that the above-mentioned embodiments are merely exemplary embodiments of the present application, and are not intended to limit the scope of the present application, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present application should be included in the scope of the present application.
Claims (13)
1. A method for predicting postoperative lung complication probability based on sleep stage data before operation is characterized by comprising the following steps:
acquiring a pre-operation physiological clinical characteristic data set of a predicted person, wherein the pre-operation physiological clinical characteristic data set comprises a pre-operation physiological characteristic data set and a pre-operation clinical characteristic data set;
acquiring the pre-operative clinical characteristic data set comprises acquiring continuous physiological signals of the sleep stage of the predicted person, wherein the continuous physiological signals comprise: continuous single lead electrocardiosignals, continuous thoracoabdominal respiration signals and continuous sleep state signals;
extracting pre-operative clinical feature data of the predicted person based on the continuous physiological signals, wherein the pre-operative clinical feature data comprises a first feature data set of heart rate variability obtained at least based on NN intervals obtained by calculation of the continuous single-lead electrocardiosignals, a second feature data set obtained at least based on the continuous thoracoabdominal respiration signals and a third feature data set obtained at least based on continuous sleep state signals;
predicting a function based on at least one feature data of each of the physiological feature data set, the first feature data set, the second feature data set and the third feature data set
Obtaining a predicted post-operative complication probability of the lung; wherein the content of the first and second substances,
，
，
array formed for input characteristic data of predicted personxThe probability of a complication of (2),θa vector formed for the coefficients of the feature,
is a constant number of times, and is,
is the characteristic coefficient corresponding to the n-th item of characteristic data,xa vector formed for the feature data,
，
is the nth characteristic data.
2. The method of claim 1, wherein the characteristic data of the first characteristic data set is: the number of preceding and following NN intervals being greater than 50ms, the average of the entire NN intervals, the high frequency energy, the ratio of low to high frequency, and the arrhythmia load.
3. The method of claim 1, wherein the characteristic data of the second characteristic data set is: mean minute ventilation during sleep, mean breathing rate during sleep, and mean inspiratory time during sleep.
4. The method of claim 1, wherein the characteristic data in the third characteristic data group calculated based on the sleep state signal is: the percentage of the rapid eye movement period REM sleep time length to the whole sleep time length, the deep sleep time length ratio and the effective blood oxygen time length ratio.
5. The method of claim 1, wherein the characteristic data in the pre-operative physiological characteristic data set is: surgical procedure data, age data, and preoperative pulmonary artery diameter data.
6. The method of claim 1, wherein the prediction function is determined by:
constructing a first predictive model from a feature data set associated with a pulmonary postoperative complication, the feature data set having an array of M feature data;
obtaining an optimal characteristic coefficient of the first prediction model through iteration;
arranging the optimal characteristic coefficients of the first prediction model according to the size sequence, and deleting the minimum characteristic coefficient and the characteristic data item corresponding to the minimum characteristic coefficient;
circularly carrying out the steps of construction, iteration, arrangement and deletion of a re-prediction model based on the residual characteristic data until the complication prediction value of the Qth prediction model is reduced by a first threshold value after a certain characteristic data is deleted;
and reserving the M-Q item feature data and the feature coefficients corresponding to the reserved feature data as final feature data and final feature coefficients respectively to obtain the prediction function.
7. The method of claim 6, wherein the probability of postoperative pulmonary complications is predicted based on sleep stage data,
the prediction model comprises a logistic regression model, and the likelihood function of the dataset based on the logistic regression model is expressed as:
wherein m is the number of continuous physiological clinical parameter arrays in the data set,
is the ith continuous physiological clinical parameter array,
is composed of
The corresponding label is marked with a corresponding label,Yis 0 or 1.
8. The method of claim 7, wherein the probability of postoperative pulmonary complications based on the sleep stage data is predicted,
the cost function based on the likelihood function is expressed as:
wherein m is the number of continuous physiological clinical parameter arrays in the data set,
is the ith continuous physiological clinical parameter array,
is composed of
The corresponding label is marked with a corresponding label,Yis 0 or 1.
9. The method of claim 8, wherein the probability of postoperative pulmonary complications is predicted based on sleep stage data,
initializing the characteristic coefficients through gradient descent and gradually updating until the characteristic coefficients optimal for the characteristic data in the continuous physiological clinical parameter array are obtained
Expressed as:
wherein the content of the first and second substances,
is as followsjThe feature coefficient corresponding to the item feature data, alpha is the learning rate,
as a cost function.
10. The method of claim 6, wherein the first threshold comprises 10% or more of the probability of postoperative pulmonary complications based on sleep stage data.
11. The method of claim 1, wherein the characteristic data extracted from the respiratory signal is obtained by performing a smoothing filtering process on the obtained respiratory signal, and removing outliers and then detecting peaks and troughs.
12. The method of claim 1, wherein the NN interval derived heart rate variability feature data is derived from an acquired electrocardiogram signal and detecting the position of the R peak.
13. The method of claim 6, wherein the forming of the prediction function further comprises: validating the predictive model, including: averagely dividing the data set into 5 parts, namely D1, D2, D3, D4 and D5, taking the D1-D4 as a verification set, and taking the D5 as the verification set, and calculating a first-folding performance parameter; sequentially circulating for 4 times, and calculating performance parameters from the second folding to the fifth folding; a degree of accuracy of the prediction function is calculated based on the performance parameter to determine whether the prediction function is appropriate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310096690.4A CN115775630A (en) | 2023-02-10 | 2023-02-10 | Postoperative lung complication probability prediction method based on sleep stage data before operation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310096690.4A CN115775630A (en) | 2023-02-10 | 2023-02-10 | Postoperative lung complication probability prediction method based on sleep stage data before operation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115775630A true CN115775630A (en) | 2023-03-10 |
Family
ID=85393485
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310096690.4A Pending CN115775630A (en) | 2023-02-10 | 2023-02-10 | Postoperative lung complication probability prediction method based on sleep stage data before operation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115775630A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117116488A (en) * | 2023-10-23 | 2023-11-24 | 北京肿瘤医院(北京大学肿瘤医院) | Method and related equipment for evaluating postoperative serious complications of aged lung cancer patients |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050267362A1 (en) * | 2004-05-17 | 2005-12-01 | Beth Israel Deaconess Medical Center | Assessment of sleep quality and sleep disordered breathing based on cardiopulmonary coupling |
| CN109833031A (en) * | 2019-03-12 | 2019-06-04 | 西安交通大学 | It is a kind of that the sleep mode automatically method by stages of more physiological signals is utilized based on LSTM |
| CN110197706A (en) * | 2019-04-26 | 2019-09-03 | 深圳市宁远科技股份有限公司 | A kind of stratification feature selection approach, system and application based on SBS |
| CN110634571A (en) * | 2019-09-20 | 2019-12-31 | 四川省人民医院 | Prognosis prediction system after liver transplantation |
| CN111407262A (en) * | 2019-04-26 | 2020-07-14 | 北京海思瑞格科技有限公司 | Sleep staging method based on B L STM and device for sleep staging based on B L STM |
| CN111417337A (en) * | 2017-12-27 | 2020-07-14 | 欧姆龙健康医疗事业株式会社 | Disease onset risk prediction device, method, and program |
| CN111860940A (en) * | 2020-06-02 | 2020-10-30 | 山西潞安环保能源开发股份有限公司 | Coal mine distribution network power failure accident prediction method based on logistic regression algorithm |
| CN113707295A (en) * | 2021-08-24 | 2021-11-26 | 中山大学附属第三医院（中山大学肝脏病医院） | Prediction method and system for senile postoperative systemic inflammatory response syndrome |
| CN114376564A (en) * | 2021-12-29 | 2022-04-22 | 华南理工大学 | Sleep staging method, system, device and medium based on cardiac shock signal |
-
2023
- 2023-02-10 CN CN202310096690.4A patent/CN115775630A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050267362A1 (en) * | 2004-05-17 | 2005-12-01 | Beth Israel Deaconess Medical Center | Assessment of sleep quality and sleep disordered breathing based on cardiopulmonary coupling |
| CN111417337A (en) * | 2017-12-27 | 2020-07-14 | 欧姆龙健康医疗事业株式会社 | Disease onset risk prediction device, method, and program |
| CN109833031A (en) * | 2019-03-12 | 2019-06-04 | 西安交通大学 | It is a kind of that the sleep mode automatically method by stages of more physiological signals is utilized based on LSTM |
| CN110197706A (en) * | 2019-04-26 | 2019-09-03 | 深圳市宁远科技股份有限公司 | A kind of stratification feature selection approach, system and application based on SBS |
| CN111407262A (en) * | 2019-04-26 | 2020-07-14 | 北京海思瑞格科技有限公司 | Sleep staging method based on B L STM and device for sleep staging based on B L STM |
| CN110634571A (en) * | 2019-09-20 | 2019-12-31 | 四川省人民医院 | Prognosis prediction system after liver transplantation |
| CN111860940A (en) * | 2020-06-02 | 2020-10-30 | 山西潞安环保能源开发股份有限公司 | Coal mine distribution network power failure accident prediction method based on logistic regression algorithm |
| CN113707295A (en) * | 2021-08-24 | 2021-11-26 | 中山大学附属第三医院（中山大学肝脏病医院） | Prediction method and system for senile postoperative systemic inflammatory response syndrome |
| CN114376564A (en) * | 2021-12-29 | 2022-04-22 | 华南理工大学 | Sleep staging method, system, device and medium based on cardiac shock signal |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117116488A (en) * | 2023-10-23 | 2023-11-24 | 北京肿瘤医院(北京大学肿瘤医院) | Method and related equipment for evaluating postoperative serious complications of aged lung cancer patients |
| CN117116488B (en) * | 2023-10-23 | 2024-01-02 | 北京肿瘤医院(北京大学肿瘤医院) | Method and related equipment for evaluating postoperative serious complications of aged lung cancer patients |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6567728B2 (en) | How to predict patient survival | |
| TWI596600B (en) | Method and system for recognizing physiological sound | |
| KR101712974B1 (en) | Apparatus, system and method for chronic disease monitoring | |
| Mikhno et al. | Prediction of extubation failure for neonates with respiratory distress syndrome using the MIMIC-II clinical database | |
| KR20200079676A (en) | Apparatus and method for inspecting sleep disorder based on deep-learning | |
| CN111407262A (en) | Sleep staging method based on B L STM and device for sleep staging based on B L STM | |
| CN110123274A (en) | A kind of monitoring system of septicopyemia | |
| CN115775630A (en) | Postoperative lung complication probability prediction method based on sleep stage data before operation | |
| Rahman et al. | Automated assessment of pulmonary patients using heart rate variability from everyday wearables | |
| CN114420301B (en) | Method, system and storage medium for predicting blood glucose based on segmented domain RF modeling | |
| CN115775629B (en) | Method for assessing risk of PPCs (point-to-point) based on sub-polar motion test before operation | |
| Arnold et al. | Estimation of airway obstruction using oximeter plethysmograph waveform data | |
| Yeginer et al. | Using lung sounds in classification of pulmonary diseases according to respiratory subphases | |
| KR102310888B1 (en) | Methods for providing information of mortality risk and devices for providing information of mortality risk using the same | |
| Wosiak et al. | On integrating clustering and statistical analysis for supporting cardiovascular disease diagnosis | |
| CN113066547B (en) | ARDS early dynamic early warning method and system based on conventional noninvasive parameters | |
| Lv et al. | Mcfn: A multichannel fusion network for sleep apnea syndrome detection | |
| US9402571B2 (en) | Biological tissue function analysis | |
| TW202302032A (en) | A method of monitoring apnea and hypopnea events based on the classification of the descent rate of heartbeat intervals | |
| Xie et al. | An improved approach for real-time detection of sleep apnea. | |
| Ravelo-García et al. | Application of RR series and oximetry to a statistical classifier for the detection of sleep apnoea/hypopnoea | |
| Maresh et al. | Role of automated detection of respiratory related heart rate changes in the diagnosis of sleep disordered breathing | |
| CN117275726A (en) | OSA (OSA) incidence risk prediction method and device based on multiple groups of biological biomarkers | |
| Petraşcu et al. | Identifying the impact of global warming on population using a clustering based approach | |
| Bao et al. | The value of small-airway function variables in spirometry, fractional exhaled nitric oxide, and circulating eosinophils for predicting bronchial hyperresponsiveness in patients with mild asthma |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |