A method for correcting potassium ion results measured by a blood gas analyzer

Abstract

The present invention belongs to the field of medical device technology, and specifically relates to a correction method for potassium ion results measured by a blood gas analyzer, which is used to solve the problem that the potassium ion results in the emergency bedside arterial blood gas analysis are always very different from the venous electrolyte potassium ion results measured by the automatic biochemical analyzer of the laboratory. The present invention collects the emergency bedside arterial blood gas results and the venous electrolyte results extracted at the same time, constructs a random forest model, and obtains the main influencing factors affecting the potassium ion results in the emergency bedside arterial blood gas analysis. According to these influencing factors, a multivariate linear regression model is fitted using multivariate linear regression. The corrected potassium ions obtained after correction by the multivariate linear regression model are basically consistent with the potassium ion results measured by the biochemical analyzer, so it can be used to correct and replace the biochemical analyzer to measure the venous blood potassium ion results, which has clear clinical significance and application promotion prospects.

Description

Technical Field

[0001] The invention relates to a correction method, in particular to a correction method for potassium ion results measured by a blood gas analyzer. Background Art

[0002] With the progress of clinical laboratory medicine and the continuous improvement of the accuracy of point-of-care testing (POCT), blood gas analysis has great advantages during surgery due to its many advantages such as convenient sampling, simple operation, easy to carry, and rapid reporting. However, the electrolyte test results of emergency bedside blood gas analyzers, especially potassium ion results, are quite different from those of fully automatic biochemical analyzers, so there are great limitations in the analysis and interpretation of clinical results. Potassium ions play an important role in diagnosing the patient's condition, especially emergency patients and intensive care patients. These patients need emergency treatment as soon as possible because of their critical condition and rapid changes. The various indicators of blood gas analysis and biochemical indicators such as electrolyte potassium, sodium, chloride, and calcium are important bases for treatment. However, the potassium ion results obtained by the blood gas analyzer are usually different from the electrolyte potassium ion results. Therefore, it is necessary to wait for the electrolyte potassium ion results from the laboratory before they can be used as the basis for diagnosis and treatment. Therefore, blood must be drawn and sent to the laboratory for testing, which often requires a long waiting time, which is not conducive to the rapid treatment of patients, and causes great inconvenience for out-of-hospital emergency treatment and incomplete equipment.

[0003] Machine learning is a computer research method that simulates or realizes human learning activities and is one of the important members in the field of artificial intelligence. In the past decade, with the rapid development of computer science and big data, solving clinical problems through machine learning algorithms based on a large amount of clinical data has become the most important development trend of smart healthcare. In 2016, Cao Wenzhe et al. used the random forest method and logistic regression method to achieve risk prediction and control of complications of stage II diabetes, and also used their ROC curve area as an evaluation indicator. In the same year, Jhajharia et al. studied the classification problem of breast cancer lesion severity based on data mining technology and machine learning methods. Compared with other common classification algorithms, the hybrid prediction model proposed in this article has the best model accuracy, which can reach 96%.

[0004] Emergency bedside arterial blood gas analysis refers to the use of a portable blood gas analyzer to obtain important parameter indicators such as pH, PCO2, PO2, electrolytes, Lac, Hb, etc. of arterial blood related to breathing, gas exchange and acid-base electrolyte balance in a very short time. It is the most reliable indicator and basis for diagnosing respiratory failure and acid-base imbalance, the most direct indicator reflecting changes in the human respiratory, circulatory and metabolic systems, and the most indispensable test item for the rescue and treatment of critically ill patients during surgery.

[0005] During surgery, arterial blood gas analysis and monitoring is an important reference for determining whether respiratory failure exists and whether various rescue and treatment measures are effective. It is of great significance to guide the adjustment of various parameters of mechanical ventilation and to correct acid-base balance and electrolyte disorders. In the body's metabolism, electrolytes play a very important role, among which potassium ions are extremely important ions, which play an irreplaceable role in the normal electrical activity of myocardial cells and the constant pH in the body. Therefore, obtaining and maintaining potassium ion results in the normal range for critically ill patients during surgery has important guiding value for the assessment of the severity of the disease, and has a very important clinical guiding significance for reducing the perioperative mortality rate of critically ill patients and ensuring the life safety of the other party of the patient during surgery. The blood gas analyzer can obtain the test report in a very short time, thereby providing a basis for the clinical treatment of the disease and providing patients with more treatment time. However, in clinical practice, it is found that the potassium ion results in the emergency bedside arterial blood gas analysis are always very different from the venous electrolyte potassium ion results measured by the automatic biochemical analyzer in the laboratory. Therefore, whether and how the arterial blood gas potassium ion results measured by the emergency bedside blood gas analyzer can replace the venous electrolyte potassium results measured by the automatic biochemical analyzer is a clinical problem and research hotspot that is urgently needed to be solved.

[0006] Based on the above background and research results, this paper corrected the arterial blood gas potassium ion results through machine learning methods, and explored the consistency of the corrected results with the electrolyte potassium ion results, aiming to provide a reference for clinical application. Summary of the invention

[0007] In view of the problems existing in the background technology, the present invention provides a method for correcting the potassium ion result measured by a blood gas analyzer, which is used to correct the result of arterial blood gas potassium ion in an arterial blood gas analyzer.

[0008] The present invention solves the above technical problems and provides the following technical solutions:

[0009] A method for correcting potassium ion results measured by a blood gas analyzer comprises the following steps:

[0010] Step 1: Collect the patient's arterial blood gas results and venous blood electrolyte potassium ion results at the same time, and use the results as processing data;

[0011] Step 2: Clean the data collected in step 1 and randomly divide the data into training set and test set according to the ratio of 70% and 30%;

[0012] Step 3: Build a random forest model based on the data processed in step 2 to obtain the top five impact factors;

[0013] Step 4: Establish a multivariate linear regression model based on the five influencing factors obtained in step 3.

[0014] After adopting this technical solution, by collecting the results of emergency bedside arterial blood gas and the results of venous electrolytes extracted at the same time, a random forest model was constructed to obtain the main influencing factors affecting the potassium ion results in emergency bedside arterial blood gas analysis, and a multivariate linear regression model was fitted based on these influencing factors. The corrected potassium ion obtained after correction by the multivariate linear regression model is basically consistent with the potassium ion result measured by the biochemical analyzer, so it can be used to correct and replace the results of venous blood potassium ion measured by the biochemical analyzer, which has clear clinical significance and application promotion prospects.

[0015] Preferably, the step 1 comprises:

[0016] All patients had venous blood collected for electrolyte testing and arterial blood collected for blood gas analysis, and venous blood was anticoagulated with heparin;

[0017] Before using the instrument for measurement, use the calibration products and supporting reagents for calibration, and perform routine maintenance on the instrument to ensure that the instrument can be used normally;

[0018] Arterial whole blood samples were tested using a blood gas analyzer, and the results of age, gender, pH, partial pressure of carbon dioxide PCO2, partial pressure of oxygen PO2, actual bicarbonate HCO3-, standard bicarbonate SBC, excess base BE, oxygen content O2Ct, blood oxygen saturation SO2%, total carbon dioxide TCO2, hematocrit Hct, alveolar oxygen partial pressure A, sodium ion Na+, chloride ion Cl-, calcium ion Ca2+, potassium ion K+, lactate Lac and oxygen concentration FIO2 were recorded;

[0019] The venous blood was centrifuged at 3000 r / min for 5 min and tested using a fully automatic biochemical analyzer to record the potassium ion K+ results.

[0020] Preferably, the step 2 further comprises statistical analysis:

[0021] For the data obtained in step 1, use Microsoft Excel 2016 to create a database for matching, deduplication and default value deletion, save it in CSV format, and import the sorted and cleaned data into RStudio;

[0022] For the data obtained in step 1, RStudio software was used to verify the normality of the data using the Shapiro-Wilk test and the homogeneity of variance using the Bartlett test.

[0023] Preferably, the step three comprises:

[0024] Use the randomForest package in the R programming language to build a random forest model, and use the psych package in the R programming language to perform pairwise variable correlation tests;

[0025] The variable sampling value mtry parameter is determined according to the average error size in the random forest model, and the number of decision trees ntree parameter is determined according to the error stability in the random forest model;

[0026] The average Gini index of the characteristic values ​​that affect the potassium ion results in arterial blood gas analysis is fitted by a random forest model, and the characteristic values ​​are sorted to select the top five influencing factors among the characteristic values;

[0027] Preferably, the step 4 comprises:

[0028] Based on the five influencing factors obtained in step 3 and the electrolyte potassium ion results obtained in step 1, a scatter plot was made using the R programming language;

[0029] According to the relationship trend between the electrolyte potassium ion results and each influencing factor, determine the relationship between each influencing factor and the electrolyte potassium ion results;

[0030] Convert the nonlinear relationship between each influencing factor and electrolyte potassium ion results into a linear relationship;

[0031] Based on the linear relationship and the transformed linear relationship between each influencing factor and the electrolyte potassium ion results, a multivariate linear regression model was fitted;

[0032] The optimal multiple linear regression model was obtained by using the stepwise regression function algorithm of the R programming language and optimizing by backward elimination method.

[0033] After adopting this preferred solution, the nonlinear relationship includes exponential, logarithmic, square, power and Boolean relationship; the backward elimination method includes: starting from the full variable regression equation, gradually deleting a variable until the index value reaches the optimal value.

[0034] Preferably, the present invention further comprises a step five following the step four:

[0035] The Shapiro-Wilk test was used to analyze whether the potassium ion results conformed to the normal distribution. If they conformed to the normal distribution, the t-test was used to test the corrected blood gas potassium ion results of the test set with the electrolyte potassium ion results of the test set to verify the effect of the multivariate linear regression model. If they did not conform to the normal distribution, the Wilcox rank sum test was used for verification.

[0036] After adopting this preferred solution, the data of the test set are verified by using a t-test or a Wilcox rank hypothesis test to determine whether the corrected potassium ion concentration data and the corresponding electrolyte potassium ion concentration data are statistically different. The closer p is to 0, the greater the difference. If p is equal to 0, they are completely different. The closer p is to 1, the smaller the difference. If p is equal to 1, they are completely the same. In this way, the similarity between the corrected potassium ion concentration obtained by the multivariate linear regression model and the actual potassium ion concentration is determined, and the validity of the multivariate linear regression model is determined.

[0037] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0038] 1. The use of emergency bedside blood gas analyzer to measure arterial blood potassium ions in combination with the correction model can achieve a high degree of consistency with the test results of laboratory equipment.

[0039] 2. A large number of relevant studies have shown that bias is inevitable in testing, which is related to many factors such as sample collection method, sample storage time, sample transportation method, sample testing equipment, and sample testing personnel experience. In addition, the anticoagulant on the inner wall of the syringe and test tube used in arterial blood collection may also cause bias. The error of the correction result is evenly distributed, that is, the frequency of high and low is similar. A result with smaller error can be obtained by averaging the correction results two or three times.

[0040] 3. The corrected potassium ions obtained by correcting K, Cl, Lac, PH, and PCO2 in blood gas analysis through the correction formula are basically consistent with the potassium ion results measured by the biochemical analyzer. Therefore, the emergency bedside blood gas analyzer can be used to measure arterial blood potassium ions in combination with the correction model to replace the venous blood potassium ion results measured by the biochemical analyzer, which has clear clinical significance and application promotion prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of the present invention.

[0042] Figure 2 This is a graph of the average error value of the random forest model as the sampling values ​​of the variables change.

[0043] Figure 3 Error plot for the random forest model augmented with decision trees.

[0044] Figure 4 Ordination plot of the average Gini index fitted for a random forest.

[0045] Figure 5 This is a variable correlation diagram.

[0046] Figure 6 Schematic diagram of the multiple linear regression model. DETAILED DESCRIPTION

[0047] In order to make the technical means, features and effects achieved by the present invention easier to understand, the technical solutions in the embodiments of the present invention are clearly and completely described below in conjunction with specific embodiments and the drawings in the embodiments of the present invention.

[0048] like Figure 1-Figure 6 As shown, the present invention provides the following technical solutions:

[0049] A method for correcting potassium ion results measured by a blood gas analyzer comprises the following steps:

[0050] Step 1: Collect the patient's arterial blood gas results and venous blood electrolyte potassium ion results at the same time, and use the results as processing data;

[0051] Step 2: Clean the data collected in step 1 and randomly divide the data into training set and test set according to the ratio of 70% and 30%;

[0052] Step 3: Build a random forest model based on the data processed in step 2 to obtain the top five impact factors;

[0053] Step 4: Establish a multivariate linear regression model based on the five influencing factors obtained in step 3.

[0054] In this embodiment, the step 1 includes:

[0055] Sample collection was based on the relevant methods for blood gas analysis in the National Clinical Laboratory Operation Procedures. All patients had venous blood collected for electrolyte testing and arterial blood collected for blood gas analysis, and heparin was used for anticoagulation.

[0056] Roche Cobas B123 blood gas analyzer and OLYMPUSAU5400 fully automatic biochemical analyzer were used, and all reagents were original reagents corresponding to the instruments;

[0057] Before using the instrument for measurement, use the calibration products and supporting reagents for calibration, and perform routine maintenance on the instrument to ensure that the instrument can be used normally;

[0058] Arterial whole blood samples were tested using Roche CobasB123 blood gas analyzer, and the results of age, gender, pH, partial pressure of carbon dioxide PCO2, partial pressure of oxygen PO2, actual bicarbonate HCO3-, standard bicarbonate SBC, excess base BE, oxygen content O2Ct, blood oxygen saturation SO2%, total carbon dioxide TCO2, hematocrit Hct, alveolar oxygen partial pressure A, sodium ion Na+, chloride ion Cl-, calcium ion Ca2+, potassium ion K+, lactate Lac and oxygen concentration FIO2 were recorded;

[0059] The venous blood was centrifuged at 3000 r / min for 5 min and tested using an OLYMPUSAU5400 fully automatic biochemical analyzer to record the potassium ion K+ concentration results.

[0060] In this embodiment, the step 2 further includes statistical analysis:

[0061] For the data obtained in step 1, use Microsoft Excel 2016 to create a database for matching, deduplication and default value deletion, save it in CSV format, and import the sorted and cleaned data into RStudio;

[0062] For the data obtained in step 1, R and RStudio software were used to verify the normality of the data using the Shapiro-Wilk test and the homogeneity of variance using the Bartlett test.

[0063] In this embodiment, the step three includes:

[0064] Use the randomForest package in the R programming language to build a random forest model, and use the psych package in the R programming language to perform pairwise variable correlation tests;

[0065] The variable sampling value mtry parameter is determined according to the average error size in the random forest model, and the number of decision trees ntree parameter is determined according to the error stability in the random forest model;

[0066] The random forest model was used to fit the average Gini index of the eigenvalues ​​that affect the potassium ion results in emergency bedside arterial blood gas analysis, and the eigenvalues ​​were sorted to select the top five influencing factors among the eigenvalues.

[0067] In this embodiment, the step 4 includes:

[0068] Based on the five influencing factors obtained in step 3 and the electrolyte potassium ion results obtained in step 1, a scatter plot was made using the R programming language;

[0069] According to the relationship trend between the electrolyte potassium ion results and each influencing factor, determine the relationship between each influencing factor and the electrolyte potassium ion results;

[0070] Convert the nonlinear relationship between each influencing factor and electrolyte potassium ion results into a linear relationship;

[0071] Based on the linear relationship and the transformed linear relationship between each influencing factor and the electrolyte potassium ion results, a multivariate linear regression model was fitted;

[0072] The optimal multiple linear regression model was obtained by using the stepwise regression function algorithm of the R programming language and optimizing by backward elimination method.

[0073] In this embodiment, step five is also included to follow step four:

[0074] The Shapiro-Wilk test was used to analyze whether the potassium ion results conformed to the normal distribution. If they conformed to the normal distribution, the t-test was used to test the corrected blood gas potassium ion results of the test set with the electrolyte potassium ion results of the test set to verify the effect of the multivariate linear regression model. If they did not conform to the normal distribution, the Wilcox rank sum test was used for verification.

[0075] The following further describes the use process of the invention in conjunction with the accompanying drawings so that those skilled in the art can fully understand the invention, which is specifically described as follows:

[0076] According to the relevant methods for blood gas analysis specimens in the "National Clinical Laboratory Operation Procedures". The blood collection subjects were mainly 2957 patients in the laboratory department of our hospital from January 1, 2021 to December 30, 2021, and the arterial blood gas results and venous electrolyte results of the same day, including 1313 females and 1644 males, aged 0 to 100 years old, with an average age of (70.50±16.52) years old. All patients had venous blood collected for electrolytes and arterial blood collected for blood gas analysis on the same day, and heparin was used for anticoagulation.

[0077] Roche Cobas B123 blood gas analyzer and OLYMPUSAU5400 fully automatic biochemical analyzer were used, and all reagents were original reagents corresponding to the instruments.

[0078] Before using the instrument for measurement, it is necessary to use calibration products and supporting reagents for calibration, and perform routine maintenance on the instrument to ensure that the instrument can be used normally; arterial whole blood samples are tested using Roche CobasB123 blood gas analyzer, and the results of age, gender, pH, partial pressure of carbon dioxide (PCO2), partial pressure of oxygen (PO2), actual bicarbonate (HCO3-), standard bicarbonate (SBC), excess base (BE), oxygen content (O2Ct), blood oxygen saturation (SO2%), total carbon dioxide (TCO2), hematocrit (Hct), alveolar oxygen partial pressure (A), sodium ion (Na+), chloride ion (Cl-), calcium ion (Ca2+), potassium ion (K+), lactate (Lac), oxygen uptake concentration (FIO2) and other results are recorded; venous blood is tested using OLYMPUSAU5400 fully automatic biochemical analyzer, centrifuged for 5 minutes at 3000r / min, and the potassium ion (K+) results are recorded.

[0079] Microsoft office Excel 2016 was used to perform data cleaning such as matching, deduplication, and default value deletion on the original data, and the data was saved in CSV format. The cleaned data was then imported into RStudio. Before building a machine learning model, the data set was randomly divided into a training set and a test set at 70% and 30% respectively.

[0080] This study selected Random Forest (RF) as a machine learning method for dealing with regression problems. First, the randomForest package in the R programming language was used to model RF, and the psych package was used to test the correlation between two variables. Then, the parameter value of the variable sampling value (mtry) was determined according to the average error size within the model, and the parameter value of the number of decision trees (ntree) was determined according to the stability of the error within the model. Then, the RF model was used to fit the characteristic values ​​(average Gini index) that affect the potassium ion results in emergency bedside arterial blood gas analysis, and these characteristic values ​​were sorted, and the top 5 factors were selected as the main factors causing the difference between blood gas potassium and electrolyte potassium ions. Finally, the prediction accuracy of the model was clarified using the test set. If it was less than 90%, the two parameters of mtry and ntree were readjusted.

[0081] Based on these five factors, a multivariate linear regression model was established, and the first correction model of arterial blood gas potassium ion was preliminarily fitted. 2 ) to evaluate the model performance, and then optimize the model according to the characteristics of each factor. By converting some linear relationships into nonlinear relationships, such as exponential, logarithmic, square, power, etc., and converting some numerical types into binary indicators, etc., a stepwise regression function is added to gradually optimize the arterial blood gas potassium ion correction model, and finally a corrected potassium model is obtained. The W test is used to analyze whether the potassium ion results conform to the normal distribution. If they conform to the normal distribution, the t test is used to test the blood gas potassium ion results after correction of the test set with the electrolyte potassium ion results of the test set to verify the effect of the machine learning correction model. If it does not conform to the normal distribution, the Wilcox rank sum test is used to verify it.

[0082] There was no statistically significant difference in gender between the training set and the test set (χ 2 =0.1232, P=0.7255), see Table 1. There was no significant difference in age between the two groups (t=-0.1443, P=0.8853)

[0083] Table 1 Gender comparison

[0084]

[0085] Table 2 Age comparison

[0086]

[0087] *Shapiro-Wilk test showed that the ages of the two groups were normally distributed, P = 0.0000

[0088] General basic information, such as: gender (Gender), age (Age), arterial blood gas analysis results, such as: PH, PCO2, PO2, HCO3-, SBC, BE, O2Ct, SO2%, TCO2, Hct, A, Na+, Cl-, Ca2+, K+, Lac, FIO2, venous electrolyte potassium ion results, with the comparison of venous electrolyte potassium ion results and arterial blood gas potassium ion as split labels. See Figure 2 As shown, the number of items with the smallest average error (0.2519829) in the model is 18 as the mtry parameter value, see Figure 3 As shown in Figure 3, the number of decision trees with stable errors in the model is about 400, which is taken as the parameter value of ntree. Then, the average Gini index affecting potassium ion results in emergency bedside arterial blood gas analysis is calculated by RF model (see Table 3 for details), and these characteristic values ​​are ranked (see Figure 4 ), the top five factors were selected, potassium (K), chloride (Cl), lactic acid (Lac), pH (PH), and partial pressure of carbon dioxide (PCO2) as the main influencing factors causing the differences between blood gas potassium and electrolyte potassium ion results.

[0089] Table 3 The average Gini index affecting potassium ion results in emergency bedside arterial blood gas analysis calculated by the RF model:

[0090]

[0091] Based on these five factors, a multivariate linear regression model was established: KJ = a + b*K.Q + c*PH + d*Cl + e*PCO2 + f*Lac. The first correction model of arterial blood gas potassium ion was preliminarily fitted. R2 was 0.6955, P = 0.0000. However, the coefficient of lactate was t = 0.653, P = 0.514, and it was considered that there was still room for optimization of the model. See Figure 5As shown in the figure, a scatter plot matrix is ​​used to understand the trend relationship between the electrolyte potassium ion results and the five influencing factors, so as to facilitate the addition of nonlinear factors. Through the scatter plot matrix, it can be roughly inferred that the electrolyte potassium ion results (KD) are logarithmic (lnX) with pH (PH), chloride ions (Cl) and carbon dioxide partial pressure (PCO2), and are polynomial (X2+…+Xn) with blood gas potassium (KQ) and lactic acid (Lac). Therefore, the optimized model is KJ=a+b*K.Q+c*PH+d*ln(Cl)+e*ln(PCO2)+f*Lac+g*Lac^2+h*Lac^3+i*Lac^4+j*Cl+k*PCO2+l*ln(PCO2)+m*KQ^2+n*KQ^3+o*KQ^4. The second correction model is obtained, R 2 The model was optimized compared with the first model, but the coefficients of Lac, PCO2, Lac^2, etc. P>0.05, so it is necessary to continue to optimize the model. Use the step function for stepwise regression and continue to optimize the model to obtain the final correction model, KJ = -239.1-3.349*KQ-35.63*PH+1.451*KQ^2-0.2175*KQ^3+0.01182*KQ^4+0.000332*Lac^3-0.00001909*Lac^4+256*ln(PH)-0.8658*ln(Cl)-0.1798*ln(PCO2), R 2 is 0.801, P = 0.0000, and all coefficients P are < 0.01, see Figure 6 As shown, the model parameters are repeatedly changed and adjusted, R 2 It is difficult to exceed 0.72, so the model is considered to be optimized.

[0092] All potassium ion results in the training set and test set were in accordance with normal distribution by Shapiro-Wilk test (P=0.0000), and Bartlett test showed homogeneity of variance (P>0.05). The average level of arterial blood gas potassium ion (KQ) of all patients was higher than the average level of venous electrolyte potassium ion (KD), and the difference was statistically significant (P<0.05). The difference between the potassium ion levels of the two groups was (0.142±0.09, 0.144±0.019), respectively, as shown in Table 4. The difference between the average level of corrected potassium ion (KJ) in the training set and test set and the average level of venous electrolyte potassium ion (KD) was (0±0.101, 0±0.107), respectively. The results of KJ and KD were basically consistent, and there was no statistically significant difference (t=0, P=1), as shown in Table 5. This shows that although the serum potassium ion results of arterial blood gas specimens are very different from the venous electrolyte potassium ion results, the correction model can be used to deduce results that are basically consistent with the venous electrolyte potassium ion results. Therefore, emergency bedside blood gas analysis can be used in conjunction with the correction model to quickly and accurately determine whether the patient's blood potassium is abnormal.

[0093] Table 4 Comparison of blood gas and electrolyte potassium ion levels in training and test sets

[0094]

[0095] *Shapiro-Wilk test showed that the potassium ion results of the two groups were in accordance with normal distribution, P = 0.0000;

[0096] Bartlett's test showed that the potassium ion variances of the two groups were equal, P>0.05

[0097] Table 5 Comparison of corrected potassium ion and electrolyte potassium ion levels in training and test sets

[0098]

[0099] *Shapiro-Wilk test showed that the potassium ion results of the two groups were in accordance with normal distribution, P = 0.0000;

[0100] Bartlett's test showed that the potassium ion variances of the two groups were equal, P>0.05

[0101] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A method for correcting potassium ion results measured by a blood gas analyzer, characterized in that: The following steps are involved: Step 1: Collect the patient's arterial blood gas results and venous blood electrolyte potassium ion results at the same time, and use the results as processing data; Step 2: Clean the data collected in step 1 and randomly divide the data into training set and test set in a ratio of 70% and 30%; Step 3: Build a random forest model based on the data processed in step 2 to obtain the top five impact factors; Step 4: Establish a multiple linear regression model based on the five influencing factors obtained in step 3; Wherein, the step one comprises: All patients had their venous blood collected for electrolyte testing and arterial blood collected for blood gas analysis. Venous blood was anticoagulated with heparin. Before using the instrument for measurement, calibration products and supporting reagents were used for calibration, and routine maintenance was performed on the instrument to ensure that the instrument could be used normally. Arterial whole blood samples were tested using a blood gas analyzer, and the results of age, gender, PH, carbon dioxide partial pressure PCO2, oxygen partial pressure PO2, actual bicarbonate HCO3-, standard bicarbonate SBC, residual base BE, oxygen content O2Ct, blood oxygen saturation SO2%, total carbon dioxide TCO2, hematocrit Hct, alveolar oxygen partial pressure A, sodium ion Na+, chloride ion Cl-, calcium ion Ca2+, potassium ion K+, lactate Lac and oxygen uptake concentration FIO2 were recorded. Venous blood was centrifuged at 3000r / min for 5 minutes in a centrifuge and tested using a fully automatic biochemical analyzer, with potassium ion K+ results recorded. The step 2 also includes statistical analysis: For the data obtained in step 1, Microsoft Excel 2016 was used to establish a database for matching, deduplication and default value deletion, and the data was saved in CSV format. The cleaned data was imported into RStudio. For the data obtained in step 1, RStudio software was used to verify the normality of the data using the Shapiro-Wilk test and the homogeneity of variance of the data using the Bartlett test. The step three comprises: The random forest model was built using the randomForest package in the R programming language, and the pairwise variable correlation test was performed using the psych package in the R programming language; the variable sampling value mtry parameter value was determined according to the average error size in the random forest model, and the number of decision trees ntree parameter value was determined according to the error stability in the random forest model; the average Gini index of the eigenvalues ​​affecting the potassium ion results in arterial blood gas analysis was fitted by the random forest model, and the eigenvalues ​​were sorted to select the top five influencing factors in the eigenvalues; The fourth step comprises: Based on the five influencing factors obtained in step three and the electrolyte potassium ion results obtained in step one, a scatter plot was made using the R programming language; according to the relationship trend between the electrolyte potassium ion results and each influencing factor, the relationship between each influencing factor and the electrolyte potassium ion results was determined; the nonlinear relationship between each influencing factor and the electrolyte potassium ion results was converted into a linear relationship; based on the linear relationship between each influencing factor and the electrolyte potassium ion results and the converted linear relationship, a multivariate linear regression model was fitted; using the stepwise regression function algorithm of the R programming language, the optimal multivariate linear regression model was obtained through backward elimination optimization; Also included is a step five for following the step four: The Shapiro-Wilk test was used to analyze whether the potassium ion results conformed to the normal distribution. If they conformed to the normal distribution, the t-test was used to test the corrected blood gas potassium ion results of the test set with the electrolyte potassium ion results of the test set to verify the effect of the multivariate linear regression model. If they did not conform to the normal distribution, the Wilcox rank sum test was used for verification.

Images