Perioperative massive transfusion nomogram for noncardiac surgery

Abstract

The application provides a non-cardiac surgery perioperative large-dose blood transfusion nomogram prediction method, relates to the technical field of medical auxiliary prediction model construction, and comprises the following steps: S1, determining the research object: adopting a retrospective case-control study method, searching for suitable non-cardiac surgery treatment cases and perioperative blood transfusion treatment cases for inclusion in the study. The R software is used for internal verification of the model, and the actual observed and predicted perioperative large-dose blood transfusion has good consistency, indicating that the calibration degree of the model is high. The prediction model can be used for the prediction of perioperative large-dose blood transfusion of various surgical patients, and is the largest sample in the existing research. Through the research on the influencing factors of large-dose blood transfusion, the prediction model of perioperative large-dose blood transfusion is established, which helps the clinicians to early identify the blood transfusion risk factors of the surgical patients before the operation and effectively predict the probability of large-dose blood transfusion of the patients during the perioperative period.

Description

Technical Field

[0001] This invention relates to the field of medical auxiliary prediction model construction technology, specifically a nomogram prediction method for high-dose blood transfusions during the perioperative period of non-cardiac surgery. Background Technology

[0002] Intraoperative massive hemorrhage is a common complication of surgery. When patients experience severe perioperative anemia or hemorrhagic shock due to intraoperative blood loss, timely and appropriate blood transfusion can save their lives. However, blood transfusion therapy also carries certain risks and costs, and blood products are extremely scarce and precious. Therefore, prudent and appropriate blood transfusion is very important. Patient-centered, precise, and individualized management can improve patient outcomes and quality of life, and can also effectively conserve precious blood resources. Therefore, perioperative red blood cell transfusion risk prediction for surgical patients, optimization of blood preparation processes, and rational allocation of blood resources are important aspects of individualized perioperative blood management.

[0003] Currently, there are studies both domestically and internationally reporting on factors influencing the risk of massive intraoperative bleeding or perioperative massive transfusion, such as patient age, body mass index (BMI), shock index, preoperative hemoglobin, left ventricular ejection fraction, and preoperative international normalized ratio (INR). These factors have been clearly proven to be associated with high-dose transfusions, but it is impossible to accurately predict the patient's transfusion risk by assessing only a single independent influencing factor.

[0004] In recent years, nomograms have been widely used in clinical research to build predictive models due to their ability to integrate different univariate variables and visualize the predicted probability results of multivariate regression models. Nomogram prediction models for blood transfusions have been reported in some subspecialty surgeries such as liver surgery, neurosurgery, and hip replacement surgery, but the sample sizes are small, ranging from only 200 to 1000 cases. Furthermore, most current studies focus on patients undergoing a single surgery or a single type of surgery, and the predicted outcomes primarily focus on whether red blood cell transfusions occur during the perioperative period. Research on the amount of red blood cells transfused during the perioperative period is limited, and there is a lack of interdisciplinary, large-sample studies. Further in-depth exploration is needed to develop a prediction model that can be applied to the perioperative red blood cell transfusion risk of patients undergoing different types of surgery. Summary of the Invention

[0005] The purpose of this invention is to provide a nomogram prediction method for perioperative high-dose transfusion in non-cardiac surgeries. Internal validation of the model using R software showed excellent consistency between actual observations and predicted perioperative high-dose transfusions, indicating high model calibration. This prediction model can be used to predict perioperative high-dose transfusions in patients undergoing various surgeries and boasts the largest sample size among existing studies. By studying the influencing factors of high-dose transfusions, a prediction model for perioperative high-dose transfusions was established to help clinicians identify transfusion risk factors in surgical patients early preoperatively, effectively predict the probability of perioperative high-dose transfusions, thereby enhancing individualized perioperative blood management, guiding surgeons to more accurately implement preoperative blood matching, providing blood transfusion departments with a basis for blood matching approval, optimizing the blood matching application process, and assessing and grading patients' physical health and surgical risk levels before anesthesia. The model's six predictive factors are all readily available, demonstrating good operability in clinical applications.

[0006] To achieve the above effects, the present invention provides the following technical solution: a method for predicting high-dose blood transfusion nomograms during the perioperative period of non-cardiac surgery, comprising the following steps:

[0007] Step 1: Determine the research subjects: A retrospective case-control study method was adopted to collect suitable cases of non-cardiac surgical treatment and perioperative blood transfusion treatment for inclusion in the study.

[0008] Step 2: Search and observe indicators: Retrospectively collect the patient's medical records during hospitalization, including the patient's general information, surgical anesthesia-related information, and preoperative blood test results.

[0009] Step 3: Univariate analysis between groups: Perform univariate analysis on the collected observation indicators for comparison between groups.

[0010] Step 4: Multivariate Logistic Regression Analysis: The statistically significant influencing factors in the univariate factors between groups are included in the binary multivariate logistic regression model. The results of the multivariate regression analysis are used to screen out independent risk factors related to perioperative high-dose blood transfusions in order to construct a nomogram model.

[0011] Step 5: Use R software to create a nomogram: Import the independent influencing factors obtained from multivariate regression analysis into the function, use R software to draw a nomogram, and construct a nomogram prediction model that can predict the risk of high-dose blood transfusions during non-cardiac surgery.

[0012] Step 6: Evaluation and Validation of the Predictive Model: Plot the receiver operating characteristic (ROC) curve using R software, determine the area under the curve, measure the discrimination of the nomogram model using the C-index, plot the calibration curve, observe the relationship between the original curve, the calibration curve, and the ideal curve, and evaluate the model's calibration accuracy.

[0013] Furthermore, the collection criteria in step one include inclusion criteria and exclusion criteria. The inclusion criteria include patients who undergo non-cardiac surgery during hospitalization and patients who undergo allogeneic red blood cell transfusion during hospitalization. The exclusion criteria include patients under 15 years of age and patients with incomplete clinical data.

[0014] Furthermore, in step one, the total transfusion volume of allogeneic red blood cells during the patient's hospitalization is divided into a non-high-dose transfusion group and a high-dose transfusion group. The total transfusion volume of the non-high-dose transfusion group is <8u of allogeneic red blood cell suspension, and the total transfusion volume of the high-dose transfusion group is ≥8u of allogeneic red blood cell suspension.

[0015] Furthermore, in step two, the general information includes gender, age, weight, height, BMI, diabetes, hypertension, heart disease, history of blood transfusion, history of allergies, history of surgery, smoking habit, and alcohol consumption.

[0016] Furthermore, in step two, the surgical anesthesia-related information includes intraoperative blood transfusion volume, surgical department, surgical name, whether it is an emergency surgery, whether it is a day surgery, surgical grade, surgical category, surgical duration, intraoperative blood loss, surgical incision description, ASA anesthesia grade, and anesthesia time.

[0017] Furthermore, in step two, the preoperative blood test results include preoperative hemoglobin, albumin, white blood cell count, platelet count, INR, PT, and APTT.

[0018] Furthermore, the steps of the three-group univariate analysis include:

[0019] S301. Normality test for continuous variables: Use functions in R to test the normality of all continuous variables. The test for normality is expressed as ks.test(x, "pnorm") # P>0.05, and the test for non-normality is expressed as P≤0.05.

[0020] S302. Nonparametric tests for continuous variables: Use functions in R language to perform nonparametric tests on continuous variables of the sample. If the difference between groups is not significant, it is expressed as wilcox.test(tested variable ~ grouping variable, paired = F)#P≥0.05. If the difference between groups is significant, it is expressed as P<0.05.

[0021] S303. Chi-square test for categorical variables: Perform chi-square test on all categorical variables using functions in R language. The identification field for sample classification is represented as tab1<-table(x, md$whether it is a high-dose blood transfusion)#"md$whether it is a high-dose blood transfusion". If the difference between groups is not significant, it is represented as CrossTable(tab1, chisq=T, expected=T, fisher=T)#P≥0.05. If the difference between groups is significant, it is represented as P<0.05.

[0022] S304. Results of univariate analysis: Based on the above univariate analysis, the results of the univariate analysis are obtained.

[0023] Furthermore, in step four, the condition for statistical significance in the inter-group univariate analysis is that the variable has a p-value of less than 0.05 in the univariate analysis.

[0024] Furthermore, in step five, the line segments in the Nomogram are marked with scores to represent the degree of influence of the factor. The sum of each individual indicator is added to obtain the total score of the predicted indicator, and finally the probability of perioperative high-dose blood transfusion can be obtained.

[0025] Furthermore, in step six, the model is internally validated using R software. A validation set is established by resampling the dataset 1000 times with replacement using the Bootstrap method, and the consistency index is calculated using the Harrell-C-statistic function.

[0026] This invention provides a nomogram prediction method for perioperative high-dose transfusion in non-cardiac surgeries, which has the following beneficial effects: Internal validation of the model using R software shows a high degree of consistency between actual observations and predicted perioperative high-dose transfusions, indicating high model calibration. This prediction model can be used to predict perioperative high-dose transfusions in patients undergoing various surgeries, and it has the largest sample size among existing studies. By studying the influencing factors of high-dose transfusions, a prediction model for perioperative high-dose transfusions is established, helping clinicians identify transfusion risk factors in surgical patients early preoperatively, effectively predict the probability of perioperative high-dose transfusions, thereby enhancing individualized perioperative blood management, guiding surgeons to more accurately implement preoperative blood matching, providing blood transfusion departments with a basis for blood matching approval, optimizing the blood matching application process, assessing and classifying patients' physical health and surgical risk levels before anesthesia, and making all six predictive factors of the prediction model readily available, resulting in good operability in clinical applications. Attached Figure Description

[0027] Figure 1 This is a flowchart of the nomogram prediction method for high-dose blood transfusion in the perioperative period of non-cardiac surgery according to the present invention;

[0028] Figure 2 This is a nomogram for predicting perioperative high-dose blood transfusion in non-cardiac surgery patients using the nomogram prediction method of the present invention.

[0029] Figure 3 This is a schematic diagram illustrating the use of the perioperative high-dose transfusion nomogram prediction model in the non-cardiac surgery perioperative high-dose transfusion nomogram prediction method of the present invention.

[0030] Figure 4 The nomogram model of the present invention for predicting high-dose blood transfusion in the perioperative period of non-cardiac surgery is used to predict the ROC curve of high-dose blood transfusion in surgical patients.

[0031] Figure 5 This is a calibration curve for the nomogram model of the non-cardiac surgery perioperative high-dose blood transfusion prediction method of the present invention to predict high-dose blood transfusion in surgical patients during the perioperative period. Detailed Implementation

[0032] Please see Figure 1-5 This invention provides a technical solution: a nomogram prediction method for high-dose blood transfusions during the perioperative period of non-cardiac surgery, comprising the following steps:

[0033] Step 1: Determine the research subjects: A retrospective case-control study method was adopted to collect suitable cases of non-cardiac surgical treatment and perioperative blood transfusion treatment for inclusion in the study.

[0034] Step 2: Search and observe indicators: Retrospectively collect the patient's medical records during hospitalization, including the patient's general information, surgical anesthesia-related information, and preoperative blood test results.

[0035] Step 3: Univariate analysis between groups: Perform univariate analysis on the collected observation indicators for comparison between groups.

[0036] Step 4: Multivariate Logistic Regression Analysis: Statistically significant univariate factors among groups are included in a binary multivariate logistic regression model. The results of the multivariate regression analysis are used to screen out independent risk factors associated with perioperative high-dose blood transfusions in order to construct a nomogram model.

[0037] Step 5: Use R software to create a nomogram: Import the independent influencing factors obtained from multivariate regression analysis into the function, use R software to draw a nomogram, and construct a nomogram prediction model that can predict the risk of high-dose blood transfusions during non-cardiac surgery.

[0038] Step 6: Evaluation and Validation of the Predictive Model: Plot the receiver operating characteristic (ROC) curve using R software, determine the area under the curve, measure the discrimination of the nomogram model using the C-index, plot the calibration curve, observe the relationship between the original curve, the calibration curve, and the ideal curve, and evaluate the model's calibration accuracy.

[0039] Specifically, the collection criteria in step one include inclusion criteria and exclusion criteria. Inclusion criteria include patients who undergo non-cardiac surgery during hospitalization and patients who receive allogeneic red blood cell transfusions during hospitalization. Exclusion criteria include patients under 15 years of age and patients with incomplete clinical data.

[0040] Specifically, in step one, the total transfusion volume of allogeneic red blood cells during hospitalization is divided into a non-high-dose transfusion group and a high-dose transfusion group. The total transfusion volume of the non-high-dose transfusion group is <8 units of allogeneic red blood cell suspension, and the total transfusion volume of the high-dose transfusion group is ≥8 units of allogeneic red blood cell suspension.

[0041] Specifically, in step two, general information includes gender, age, weight, height, BMI, diabetes, hypertension, heart disease, history of blood transfusion, history of allergies, history of surgery, smoking and alcohol consumption.

[0042] Specifically, in step two, the surgical anesthesia-related information includes the amount of blood transfused during the operation, the surgical department, the name of the operation, whether it is an emergency operation or a day surgery, the surgical grade, the type of operation, the duration of the operation, the amount of blood loss during the operation, the description of the surgical incision, the ASA anesthesia grade, and the anesthesia time.

[0043] Specifically, in step two, the preoperative blood test results include preoperative hemoglobin, albumin, white blood cell count, platelet count, INR, PT, and APTT.

[0044] Specifically, the steps of the three-group univariate analysis include:

[0045] S301. Normality test for continuous variables: Use functions in R to test the normality of all continuous variables. The test for normality is expressed as ks.test(x, "pnorm") # P>0.05, and the test for non-normality is expressed as P≤0.05.

[0046] S302. Nonparametric tests for continuous variables: Use functions in R language to perform nonparametric tests on continuous variables of the sample. If the difference between groups is not significant, it is expressed as wilcox.test(tested variable ~ grouping variable, paired = F)#P≥0.05. If the difference between groups is significant, it is expressed as P<0.05.

[0047] S303. Chi-square test for categorical variables: Perform chi-square test on all categorical variables using functions in R language. The identification field for sample classification is represented as tab1<-table(x, md$whether it is a high-dose blood transfusion)#"md$whether it is a high-dose blood transfusion". If the difference between groups is not significant, it is represented as CrossTable(tab1, chisq=T, expected=T, fisher=T)#P≥0.05. If the difference between groups is significant, it is represented as P<0.05.

[0048] S304. Results of univariate analysis: Based on the above univariate analysis, the results of the univariate analysis are obtained.

[0049] Specifically, in step four, the condition for statistical significance in univariate analysis between groups is that the variable has a p-value less than 0.05 in the univariate analysis.

[0050] Specifically, in step five, the line segments in the Nomogram are marked with scores to represent the degree of influence of the factor. The sum of each individual indicator is added together to obtain the total score of the predicted indicator, and finally the probability of perioperative high-dose blood transfusion can be obtained.

[0051] Specifically, in step six, the model is internally validated using R software. A validation set is established by resampling the dataset 1000 times with replacement using the Bootstrap method, and the consistency index is calculated using the Harrell-C-statistic function.

[0052] The method described in the embodiments was analyzed and compared with existing technologies. It was found that when the model was internally validated using R software, there was a good consistency between the actual observed and predicted perioperative high-dose blood transfusions, indicating a high degree of model calibration. This predictive model can be used to predict perioperative high-dose blood transfusions for various surgical patients and has the largest sample size among existing studies. By studying the influencing factors of high-dose blood transfusions, a predictive model for perioperative high-dose blood transfusions was established to help clinicians identify transfusion risk factors in surgical patients early before surgery, effectively predict the probability of perioperative high-dose blood transfusions, thereby enhancing individualized perioperative blood management, guiding surgeons to more accurately implement preoperative blood matching, providing blood matching approval criteria for blood transfusion departments, optimizing the blood matching application process, and assessing and classifying the patient's physical health and surgical risk level before anesthesia. The model's six predictive factors are relatively easy to obtain, making it highly operable in clinical applications.

[0053] This invention provides a nomogram prediction method for high-dose blood transfusions during the perioperative period of non-cardiac surgery:

[0054] I. Materials and Methods for Model Building

[0055] 1.1 Determine the modeling objects and groupings

[0056] 1.1.1 Research Subjects

[0057] This study employed a retrospective case-control approach. Suitable cases who underwent non-cardiac surgery and received perioperative blood transfusions were included in the study.

[0058] Inclusion criteria: Patients who underwent non-cardiac surgery during hospitalization; and patients who received allogeneic red blood cell transfusion during this hospitalization.

[0059] Exclusion criteria: age less than 15 years; incomplete clinical data of the case.

[0060] 1.1.2 Experimental Grouping

[0061] Based on the total amount of allogeneic red blood cell suspension transfused during hospitalization, patients were divided into a non-high-dose transfusion group (allogeneic red blood cell suspension transfusion volume <8u) and a high-dose transfusion group (allogeneic red blood cell suspension transfusion volume ≥8u). According to the standard requirements for whole blood and blood component quality, each unit of whole blood contains 200 ml of red blood cells.

[0062] 1.2 Research Indicators

[0063] 1.2.1 Outcome Indicators:

[0064] The study used 8 units or more of allogeneic red blood cell suspension transfused during and within 24 hours after surgery as the outcome measure. This measure was also used as the basis for study grouping and as a predictive outcome for the model.

[0065] 1.2.2 Forecasting Indicators:

[0066] Retrospectively collect the patient's medical records during hospitalization, including the following three parts:

[0067] (1) General information of the patient: gender, age, weight, height, BMI [(kg / m²)] 2 = weight (kg) / height 2 m 2 [etc.]; past medical history includes: diabetes, hypertension, heart disease (coronary heart disease, heart failure, organic heart disease, etc.), history of blood transfusion, history of allergies, history of surgery, smoking, alcoholism, etc.

[0068] (2) Surgical anesthesia related information: intraoperative blood transfusion volume, surgical department, name of surgery, whether it is an emergency surgery, whether it is a day surgery, surgical grade, surgical category, duration of surgery, intraoperative blood loss, description of surgical incision, ASA anesthesia grade, anesthesia time, etc.

[0069] (3) Preoperative blood test results of the patient: preoperative hemoglobin, albumin, white blood cell count, platelet count, INR, PT, APTT, etc.

[0070] 1.3 Statistical Analysis

[0071] The main statistical work of the study was completed using R software (Version 4.3.2).

[0072] 1.3.1 Univariate analysis:

[0073] The Kolmogorov-Smirnov test was used to check whether the data for each group conformed to a normal distribution. Normally distributed continuous data were described as mean ± standard deviation, and t-tests were used for comparisons between groups. Non-normally distributed continuous data were described as median [upper quartile to lower quartile], and Wilcoxon tests were used for comparisons between groups. For count data, frequency (percentage) was used, and chi-square tests were used for comparisons between groups.

[0074] 1.3.2 Binary multivariate logistic regression analysis:

[0075] Variables with a p-value less than 0.05 in univariate analysis were considered to have statistical significance in intergroup comparisons. These variables were further introduced into a binary multivariate logistic regression model. The results of the multivariate regression analysis were used to screen out independent risk factors associated with perioperative high-dose blood transfusions in order to construct a nomogram model.

[0076] 1.3.3 Constructing a Column Chart Model

[0077] Using the `lrm` function from the `rms` package in R software, independent influencing factors obtained from multivariate regression analysis were imported into the function. A nomogram was then created using R software to construct a nomogram prediction model that can predict the risk of high-dose blood transfusions during non-cardiac surgery. In the nomogram, the line segments are marked with scores to represent the degree of influence of each factor. The sum of each individual indicator is used to obtain the total predicted score, which ultimately yields the probability of high-dose blood transfusions during the perioperative period.

[0078] 1.3.4 Validation and Evaluation of the Prediction Model

[0079] Internal validation of the model was performed using R software. A validation set was built using the Bootstrap method with 1000 resampling cycles with replacement, and the consistency index (C-index) was calculated using the Harrell-C-statistic function. Receiver operating characteristic (ROC) curves were also plotted using R software to determine the area under the curve (AUC), which, together with the C-index, measures the discriminative power of the nomogram model. Finally, the calibration plot was generated using the calibrate function in R software to observe the relationship between the original model curve, the calibration curve, and the ideal curve, thus evaluating the model's calibration accuracy.

[0080] II. Specific Implementation Steps and Results of Model Establishment

[0081] The following section further describes the establishment of this predictive model patent technology, combining specific implementation steps and corresponding results:

[0082] 1. Data collection and preprocessing

[0083] 1.1 Collection and preprocessing of research subjects:

[0084] 1.1.1 Collection of Study Subjects: Information on patients who received perioperative red blood cell suspension transfusions within a certain period was collected through the blood transfusion patient information system. According to the blood transfusion department's information system recording protocol, each collection and use of blood products generates a record in the system. Key information initially obtained included the patient's hospital number, the amount of blood drawn each time, and the date of blood draw. A list of hospital numbers for perioperative blood transfusion patients was compiled, forming the study subject list.

[0085] 1.1.2 Summary of Outcome Indicators: Using Excel's data statistics function, multiple blood collection records for each patient during a single surgery and the same perioperative period were summarized and merged. The total red blood cell dose taken by each patient during the perioperative period was statistically analyzed. The total amount of red blood cell suspension transfused (U) for each patient during the perioperative period of this surgery is the outcome indicator for establishing the model training set.

[0086] 1.1.3 Sample Grouping:

[0087] Based on the outcome measure, namely the total amount of allogeneic red blood cell suspension transfused into the patient during the perioperative period, the study subjects were divided into a non-high-dose transfusion group (allogeneic red blood cell suspension transfusion volume <8u) and a high-dose transfusion group (allogeneic red blood cell suspension transfusion volume ≥8u).

[0088] 1.2. Collection and Preprocessing of Predictors

[0089] 1.2.1 Collection of Observational Indicators for Perioperative High-Dose Transfusion: First, by reviewing existing domestic and international research reports on factors influencing the risk of massive intraoperative bleeding or perioperative high-dose transfusion, it was found that factors such as patient age, body mass index (BMI), shock index, preoperative hemoglobin, left ventricular ejection fraction, and preoperative international normalized ratio (INR) have been clearly proven to be associated with high-dose transfusion. Second, based on clinical experience, factors potentially related to perioperative transfusion were compiled, including general information such as patient gender, height, and weight; surgical type, operation duration, and anesthesia grade; and preoperative laboratory indicators such as white blood cell count and platelet count. Using the hospital's big data platform and His Inpatient Information System, the hospital admission numbers of patients in the compiled "Study Subject List" were imported, and predictive factors potentially associated with high-dose transfusion for each patient were exported from the system to obtain a database of observational indicators.

[0090] 1.2.2 Preprocessing of predictors:

[0091] (1) Clean the database and discard relevant data that cannot be used for correlation analysis:

[0092] Some predictive factors in the preliminary observation index database could not be correlated due to missing key information. For example, the exported "imaging examination" item could not be viewed for specific examination time due to data platform recording reasons, and could not be correlated with the perioperative blood transfusion time, so it was discarded. For example, some blood test items such as "arterial blood gas analysis results" are not routine perioperative examination items, and only some patients have this data, so they cannot be statistically analyzed and were therefore discarded.

[0093] (2) Handling missing values:

[0094] Missing values ​​refer to data gaps or truncation in coarse data caused by a lack of information, resulting in incomplete values ​​for one or more attributes in the dataset. Missing value handling solutions are as follows:

[0095] Use the `summary()` function from the `base` package in R to calculate the number of missing values, and the `which(is.na())` function to find the location of missing values.

[0096] Variables with more than 80% missing values ​​will be deleted directly.

[0097] For variables with missing values ​​between 30% and 80%, the missing values ​​of the variable are treated as a separate category for inter-group univariate analysis.

[0098] For variables with less than 30% missing values, missing values ​​are imputed. For continuous variables that follow a normal distribution, the mean is used for imputation; for continuous variables that do not follow a normal distribution, the median is used for imputation; and for categorical variables, the mode is used for imputation.

[0099] (3) Numerical transformation of continuous variables:

[0100] BMI (Body Mass Index): The BMI value for each patient is obtained by using a custom function in Excel to calculate the square of the patient's weight / height (kg / m2) from the database.

[0101] Surgery / anesthesia duration: The surgery / anesthesia duration (in minutes) is obtained by using the time difference calculation function in Excel software to calculate the end time of the patient's surgery / anesthesia and the start time of the surgery / anesthesia obtained from the database.

[0102] 2. One-way analysis between groups

[0103] 2.1 Normality test for continuous variables:

[0104] Perform normality tests on all continuous variables: operation duration (min), anesthesia duration (min), height, weight, BMI, intraoperative blood loss, preoperative hemoglobin level, preoperative albumin level, preoperative white blood cell count, preoperative platelet count, preoperative INR, preoperative PT, and preoperative APTT using the Kolmogorov-Smirnov Tests function in the stats package of R.

[0105] ks.test(x,"pnorm") # P>0.05 indicates that the distribution follows a normal distribution; P≤0.05 indicates that the distribution does not follow a normal distribution.

[0106] The final test results indicate that none of the continuous variables conform to a normal distribution.

[0107] 2.2 Nonparametric tests for continuous variables:

[0108] Use the Wilcoxon Rank Sum and Signed Rank Tests functions from the stats package in R to perform nonparametric tests on continuous variables in the sample:

[0109] wilcox.test(tested variable ~ grouping variable, paired = F) # P ≥ 0.05 indicates no significant difference between groups; P < 0.05 indicates a significant difference between groups.

[0110] 2.3 Chi-square test was used for categorical variables:

[0111] All categorical variables—gender, preoperative anemia, day surgery, emergency surgery, surgical category, surgical grade, ASA grade, surgical incision description, history of hypertension, history of diabetes, history of heart disease, history of blood transfusion, history of allergies, surgical history, smoking history, and history of alcohol consumption—were subjected to a chi-square test using the `Cross Tabulation with Tests for Factor Independence` function from the `gmodels` package in R.

[0112] tab1 <- table(x, md$whether it is a high-dose blood transfusion) # "md$whether it is a high-dose blood transfusion" is the identification field for sample classification.

[0113] CrossTable(tab1,chisq=T,expected=T,fisher=T)# P≥0.05 indicates no significant difference between groups; P<0.05 indicates a significant difference between groups.

[0114] 2.4 Results of univariate analysis:

[0115] Based on the univariate analysis above, the results were obtained (see Table 1 for details): age, weight, BMI, preoperative albumin level, preoperative APTT, operation duration, anesthesia duration, surgical grade, ASA grade, surgical incision description, history of hypertension, surgical history, and intraoperative blood loss were associated with perioperative high-dose blood transfusion (p < 0.05, statistically significant differences between groups). Height, preoperative hemoglobin level, preoperative white blood cell count, preoperative INR, preoperative PT, preoperative platelet count, gender, whether it was a day surgery, whether it was an emergency surgery, surgical category, diabetes, heart disease, history of blood transfusion, history of allergies, smoking history, and alcohol consumption history were not significantly associated with perioperative high-dose blood transfusion (p > 0.05, no statistically significant differences between groups).

[0116] Table 1. Results of univariate analysis of factors influencing perioperative high-dose blood transfusion

[0117]

[0118]

[0119]

[0120]

[0121] II. Binary Multivariate Logistic Regression Analysis

[0122] Before multivariate analysis, considering the ease of use of the regression model, the univariate variables were assigned values ​​as categorical variables according to their respective clinical significance or clinical usage habits:

[0123] Table 2. Assignment table of factors related to perioperative high-dose blood transfusion

[0124]

[0125] Whether the patient received perioperative high-dose blood transfusion was used as the dependent variable. Variables with statistical significance (P < 0.05) in the intergroup comparisons of univariate analysis were selected: age, operation duration, BMI, intraoperative blood loss, preoperative albumin level, APTT, preoperative anemia, surgical grade, ASA anesthesia grade, surgical incision description, history of hypertension, and previous surgical history (variable values ​​are shown in Table 2). A total of 12 factors related to perioperative high-dose blood transfusion were included in multivariate logistic regression analysis.

[0126] The Fitting Generalized Linear Models function in the stats package of R software is used to perform regression analysis on the relationship between the dependent and independent variables. The code is as follows, where "mydata$whether it is a high-dose blood transfusion" is the basis for binary classification of the sample, and "family=binomial" is the binary logistic regression analysis method selected in the regression analysis family of the package.

[0127] R code:

[0128] glm(mydata$ Is a large-dose blood transfusion possible?)

[0129] mydata$operation duration + mydata$BMI + mydata$blood loss + mydata$APTT + mydata$albumin + mydata$surgical history + mydata$anemia status + mydata$ASA classification + mydata$surgical grade, family = binomial

[0130] data = mydata)

[0131] Furthermore, the OR value and 95% confidence interval (CI) were calculated, and the results of the multivariate regression analysis were summarized:

[0132] Table 3 Multivariate Logistic Regression Analysis of Factors Influencing Perioperative High-Dose Blood Transfusion

[0133]

[0134]

[0135]

[0136] IV. Drawing a nomogram prediction model

[0137] Based on the results of multivariate logistic regression analysis, five predictor variables were included: operation duration, intraoperative blood loss, preoperative albumin level, surgical grade, and anesthesia grade. Preoperative anemia was also included as a predictor variable due to its significant clinical importance in assessing perioperative blood transfusion and blood combination. The six predictor variables were assigned values ​​according to their clinical significance and the model's practicality, and were set as categorical variables.

[0138] Using the `Logistic Regression Model` and `Draw a Nomogram Regressing a RegressionFit` functions from the `Package rms version 6.7-1` and `Package base version 4.3.1` packages in R software, a multivariate regression relationship was established between the six predictor variables and the independent variables. This multivariate regression relationship was then transformed into a nomogram model. The specific code and steps are as follows:

[0139]

[0140]

[0141]

[0142] Where lrm is the function for establishing multifactor regression relationships, nomogram is the function for converting regression relationships into nomograms, and "md" (mydata) is the data table of 5252 research samples and six variables included in the regression relationship.

[0143] Running the program yields the following results: Figure 2 The nomogram shown is a prediction of perioperative high-dose blood transfusions in surgical patients.

[0144] Each individual indicator has a corresponding score (Points). The sum of the scores of all individual indicators predicts the total score of the indicator (Total Points). The total score of the predictive indicator corresponds to the probability value on the high-dose blood transfusion line segment. The probability of perioperative high-dose blood transfusion can be obtained by reading the graph.

[0145] The following is an example of how to use this chord chart prediction model:

[0146] For example, a patient with a pelvic fracture is scheduled for open osteotomy and internal fixation. The surgery is expected to last 5 hours, with an estimated intraoperative blood loss of 1200 ml. Preoperative blood tests show albumin 28 g / L, hemoglobin (Hb) 81 g / L, surgical grade 4, and ASA grade III. During preoperative preparation, the surgeon uses a nomogram to predict the patient's perioperative transfusion risk. The individual scores for each predictive indicator are as follows: surgery duration 14, intraoperative blood loss 32, preoperative hemoglobin level 26, preoperative anemia 6, surgical grade 40, ASA grade 10, and the total predictive indicator score is 128. The predicted probability of a high-volume perioperative transfusion (more than 8 units of red blood cells) for this patient is approximately 0.2, suggesting a relatively low risk of high-volume perioperative transfusion.

[0147] 1. Evaluation of the predictive performance of this nomogram model

[0148] The model was internally validated using R software. 1000 resamplings with replacement were performed, and the Bootstrap method was used to perform internal cross-validation of the model. The mean C-index of the resampling was calculated using the Harrell C statistic function, which was 0.859, with a 95% CI [0.857, 0.861].

[0149] Furthermore, the ROC curve of this model was plotted for further analysis, and its area under the ROC curve (AUC) was calculated in R software to be 0.864 (see details). Figure 4 ).

[0150] Finally, calibration plots were generated, and it was observed that the model's original curve (apparent) and the bias-corrected curve after internal model validation closely matched the ideal curve (see details). Figure 5 The results indicate a good agreement between the observed and predicted perioperative high-dose blood transfusions, suggesting a high degree of model calibration.

[0151] 2. The innovativeness of this nomogram prediction model

[0152] 2.1 Wide range of applicable objects

[0153] Current domestic research on nomogram prediction models for blood transfusion mostly focuses on patients undergoing a single discipline or type of surgery. There is a lack of interdisciplinary nomogram prediction models that can be applied simultaneously to patients undergoing different types of surgery, covering perioperative red blood cell transfusion risk. This study collected data from patients undergoing all surgical types except cardiac surgery to screen for transfusion risk and establish a model. This prediction model can be used to predict perioperative high-dose blood transfusions in patients undergoing various surgeries.

[0154] 2.2 Large sample size

[0155] A review of relevant literature revealed that current nomogram prediction models for perioperative transfusion risk have been reported in some subspecialty surgeries such as liver surgery, neurosurgery, and hip replacement surgery, but the sample sizes for establishing nomogram regression models are small (only 200–1000 cases). This study collected data from 5252 non-cardiac surgical patients who received transfusions within the past five years, representing the largest sample size among domestic studies establishing nomogram prediction models for transfusion risk.

[0156] 2.3 For the first time, a predictive model was established for "high-dose" blood transfusions in the blood preparation process.

[0157] Current research on perioperative transfusion risk, both domestically and internationally, primarily focuses on whether red blood cell transfusion occurs in surgical patients during the perioperative period, with limited research on the amount of red blood cells transfused in surgical patients during the perioperative period.

[0158] According to the relevant core medical system, the "Clinical Blood Use Review System," when a patient's total blood usage request on the same day reaches or exceeds 1600ml (i.e., 8U of red blood cell suspension), a request must be submitted by a professional physician with intermediate or higher professional titles, approved by the department head, and then submitted to the Medical Affairs Department for final approval before blood can be prepared. A review of domestic and international literature on the risk of perioperative blood loss in non-cardiac surgeries reveals no reports of risk factors predicting perioperative blood loss exceeding 1600ml, nor are there any guidelines recommending which types of surgeries allow for preoperative blood mixing exceeding 8U. Therefore, whether preoperative blood mixing exceeds 8U is currently largely based on the surgeon's subjective judgment, lacking objective evidence and guideline recommendations. Furthermore, the Medical Affairs Department lacks a basis for approving applications for preoperative blood mixing exceeding 8U. Thus, in clinical practice, for surgical cases with potentially high expected bleeding risks, it is difficult to effectively avoid the adverse effects of insufficient blood preparation on the surgery and the patient, or the waste of blood resources due to unused blood reserves.

[0159] According to the aforementioned regulations of the "Clinical Blood Use Review System," this application innovatively defines "high-dose blood transfusion" as a blood transfusion (red blood cell suspension) volume greater than 8 units during the perioperative period. By studying the influencing factors of high-dose blood transfusion, a predictive model for perioperative high-dose blood transfusion is established to help clinicians identify blood transfusion risk factors in surgical patients in the early preoperative period, effectively predict the probability of patients undergoing high-dose blood transfusion during the perioperative period, thereby enhancing individualized perioperative blood management, guiding surgeons to implement preoperative blood matching more accurately, providing blood matching approval basis for blood transfusion departments, and optimizing the blood matching application process.

[0160] 2.4 It has good operability in clinical application.

[0161] Before using this experimental model to predict high-dose blood transfusions, it is necessary to collect six predictive indicators: operation time, intraoperative blood loss, preoperative albumin level, preoperative anemia, surgical grade, and ASA anesthesia grade. Among them, the operation time and intraoperative blood loss are modeled using relevant data from the patient's medical records and surgical records. When using the model, for different surgical subspecialties and different surgical methods, it is necessary to make predictions based on factors such as the patient's surgical site, lesion extent, surgical difficulty, and the surgeon's skill level. In the nomogram model, "operation duration" is predicted with different levels of variables: "<2h, 2-4h, 4-6h, 6-8h, >8h". "Intraoperative blood loss" is predicted with different levels of variables: "<1000ml, 1000-2000ml, 2000-3000ml, 3000-4000ml, >4000ml". This model provides feasibility and practicality for using "estimated operation duration" and "estimated intraoperative blood loss" in clinical practice. Furthermore, the final nomogram model uses "estimated operation duration" and "estimated blood loss" as subheadings for the corresponding score indicators. "Preoperative albumin level" and "preoperative anemia" can be directly and objectively determined from preoperative blood test results, providing the classification of preoperative albumin levels and anemia. This system is used to assess and classify patients' physical health and surgical risk before anesthesia. Both of these evaluation indicators are commonly used surgical risk assessment and grading indicators in preoperative evaluation by clinicians. All six predictive factors of this model are relatively easy to obtain, making it highly feasible for clinical applications.

[0162] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A nomogram prediction method for high-dose blood transfusions during the perioperative period of non-cardiac surgery, characterized in that, Includes the following steps: S1. Determine the research subjects: A retrospective case-control study method will be adopted to collect suitable cases of non-cardiac surgical treatment and perioperative blood transfusion treatment for inclusion in the study; S2. Search observation indicators: Retrospectively collect the patient's medical records during hospitalization, including the patient's general information, surgical anesthesia-related information, and preoperative blood test results; S3. Univariate analysis between groups: The collected observation indicators are compared between groups using univariate analysis. S4. Multivariate Logistic Regression Analysis: The statistically significant influencing factors in the univariate factors between groups are included in the binary multivariate logistic regression model. The results of the multivariate regression analysis are used to screen out independent risk factors related to perioperative high-dose blood transfusion in order to construct a nomogram model. S5. Use R software to build a nomogram: Import the independent influencing factors obtained from multivariate regression analysis into the function, use R software to draw a nomogram, and build a nomogram prediction model that can predict the risk of high-dose blood transfusion during non-cardiac surgery. S6. Evaluation and validation of the prediction model: Plot the receiver operating characteristic curve using R software, determine the area under the curve, measure the discrimination of the nomogram model using the C-index, plot the calibration plot, observe the relationship between the original curve, the calibration curve, and the ideal curve, and evaluate the calibration of the model. The total transfusion volume of allogeneic red blood cells during hospitalization was divided into a non-high-dose transfusion group and a high-dose transfusion group. The total transfusion volume of the non-high-dose transfusion group was <8 units of allogeneic red blood cell suspension, and the total transfusion volume of the high-dose transfusion group was ≥8 units of allogeneic red blood cell suspension. The general information includes gender, age, weight, height, BMI, diabetes, hypertension, heart disease, history of blood transfusion, history of allergies, history of surgery, smoking and alcohol consumption; The surgical anesthesia-related information includes intraoperative blood transfusion volume, surgical department, surgical name, whether it is an emergency surgery, whether it is a day surgery, surgical grade, surgical category, surgical duration, intraoperative blood loss, surgical incision description, ASA anesthesia grade, and anesthesia time.

2. The method for predicting perioperative high-dose transfusion nomograms for non-cardiac surgery according to claim 1, characterized in that, The collection criteria in step S1 include inclusion criteria and exclusion criteria. The inclusion criteria include patients who undergo non-cardiac surgery during hospitalization and patients who undergo allogeneic red blood cell transfusion during hospitalization. The exclusion criteria include patients under 15 years of age and patients with incomplete clinical data.

3. The method for predicting perioperative high-dose transfusion nomograms for non-cardiac surgery according to claim 1, characterized in that, In step S2, the preoperative blood test results include preoperative hemoglobin, albumin, white blood cell count, platelet count, INR, PT, and APTT.

4. The method for predicting perioperative high-dose transfusion nomograms for non-cardiac surgery according to claim 1, characterized in that, In step S4, the condition for statistical significance in the inter-group univariate analysis is that the variable has a p-value of less than 0.05 in the univariate analysis.

5. The method for predicting perioperative high-dose transfusion nomograms for non-cardiac surgery according to claim 1, characterized in that, In step S5, the line segments in the Nomogram are marked with scores to represent the degree of influence of the factor. The sum of each individual indicator is added to obtain the total score of the predicted indicator, and finally the probability of perioperative high-dose blood transfusion can be obtained.

6. The method for predicting perioperative high-dose transfusion nomograms for non-cardiac surgery according to claim 1, characterized in that, In step S6, the model is internally validated using R software. A validation set is established by resampling the dataset 1000 times with replacement using the Bootstrap method, and the consistency index is calculated using the Harrell-C-statistic function.

Images