CN1973778A - Method of predicting serious complication risk degree after gastric cancer operation - Google Patents

Abstract

The invention provides an objective, quantified and highly accurate method for predicting the risk of serious complications after gastric cancer surgery. The method includes the following steps: (1) through comprehensive univariate analysis, several factors that have an impact on severe complications after gastric cancer surgery are screened out; (2) through binary multiple unconditional logistic regression analysis, the determinants that really affect the prognosis are determined (3) Determine the cut-off value of the best predictive value through receiver operating characteristic curve analysis, (4) Establish a predictive model with major risk factors as independent variables, and judge the risk of severe postoperative complications of patients by predicting probability risk. The invention is a high-precision, comprehensive and objective, easy-to-operate and intelligent risk assessment system.

Description

Method for predicting the risk of serious complications after gastric cancer surgery

1. Technical field

The invention relates to a prediction method used in medicine, in particular to a method for predicting the risk of severe postoperative complications through the preoperative information, operation information, and tumor pathological information of gastric cancer patients.

2. Background technology

In my country, gastric cancer is one of the most common malignant tumors, and its morbidity and mortality rank among the forefront of various malignant tumors. The main treatment methods for gastric cancer are surgery and adjuvant chemotherapy. Radical gastric cancer surgery is currently the only treatment that can achieve the purpose of cure. The standard radical surgery for gastric cancer includes gastrectomy, lymph node dissection, and reconstruction of the digestive tract. The surgical trauma is large and the operation is complicated. Since most gastric cancer patients in my country are elderly, they often have chronic co-morbidities, and most of the tumors are middle-advanced cancers. See, even in large medical units with rich surgical experience, the incidence of serious complications after gastric cancer surgery is still as high as 10% to 20%. Common serious postoperative complications of gastric cancer are as follows: (1) severe infection, including chest infection, abdominal infection, incision infection, etc.; (2) intestinal obstruction, including acute input loop obstruction, anastomotic stoma obstruction, output loop obstruction, gastroparesis, etc. (3) Fistula, including pancreatic fistula, anastomotic fistula, duodenal stump fistula, etc.; (4) Multiple organ dysfunction, including heart, lung, liver, kidney and other organ failure, stress ulcer, DIC wait. Compared with other operations, severe postoperative complications of gastric cancer are more unique and complex. Once it occurs, the clinical treatment is extremely difficult, and the treatment costs are huge. In severe cases, multiple organ failure is easily secondary to life-threatening and death. rate as high as 24%. Therefore, it is of great clinical and social value to determine the risk factors and risk degree of severe complications after radical gastrectomy.

In the research field of risk factor determination and risk prediction of severe complications after gastric cancer surgery, there are still many shortcomings in the current research at home and abroad: (1) Some researchers collected few pathogenic factors and participated in fewer cases, which did not meet the One of the core requirements of modern evidence-based medicine, that is, clinical evidence should come from multi-center, large-sample randomized controlled clinical trials (RCT), systematic review (systematic review) and meta-analysis (meta-analysis); (2) some The researchers used a univariate analysis method, but the relationship between severe complications after gastric cancer surgery and many pathogenic factors is very complicated, and the univariate analysis cannot balance the effects of multiple confounding factors in the complex relationship, nor can it form a predictive model; (3) Some researchers use multiple linear regression analysis methods, unable to determine the optimal cut-off value of the prediction probability, unable to accurately verify and evaluate the established model, and have poor practicability and operability; severe postoperative complications and The relationship between various pathogenic factors is actually not a linear relationship. At the same time, a good prediction model must determine the best cut-off value of the prediction probability, and must pass strict verification to prove its high accuracy. Sensitivity, specificity, and requires simple operation and strong practicability; (4) To find out the significant factors from the many factors that affect the serious postoperative complications, in order to achieve the purpose of predicting the prognosis of the disease, it is necessary to analyze the traditional The prediction method is further expanded, and the knowledge and ideas of the frontier of statistics are introduced into the data analysis method.

3. Contents of the invention

The purpose of the present invention is to provide a method for predicting the risk of severe complications after gastric cancer surgery. The present invention conducts a comprehensive and detailed retrospective investigation of clinical indicators such as preoperative information, surgical information, and tumor pathological information of gastric cancer patients, and determines the severity of postoperative complications of gastric cancer through binary multivariate unconditional logistic regression (Logistic Regression) analysis. Calculate the relative risk of the main risk factors; determine the best prediction cut-off value through receiver operating characteristic curve (ROC) analysis, and evaluate the sensitivity and specificity of the prediction method; establish a prediction model with the main risk factors as independent variables , so as to provide an objective basis for the assessment of the risk of severe postoperative complications, and achieve the functions of helping patients make medical decisions, assisting ward management, and guiding medical clinical research.

In order to achieve the purpose of the present invention, the present invention provides a method for evaluating the risk of serious complications after gastric cancer surgery by using binary multiple unconditional logistic regression (LogisticRegression) analysis technology and receiver operating characteristic curve (ROC) analysis technology , the prediction method includes the following steps:

1. The SPSS 13.0 software package was used to establish a gastric cancer information database, and the recorded 79 variables were as follows: (1) Continuous variables: number of lymph node metastasis, number of lymph node dissection, intraoperative blood transfusion, intraoperative time, tumor diameter, age, Plasma albumin, prealbumin, Child-pugh score of liver function, total bilirubin, hemoglobin, white blood cell count, lymphocyte count, prothrombin time, blood sugar, carcinoembryonic antigen; (2) Ordinal variable: lymph node (LN ) degree of resection, degree of radical operation, T stage, N stage, TNM stage, degree of tumor differentiation, Borrman classification; (3) dichotomous variables: group 10 LN dissection, group 11p LN dissection, group 12 LN dissection, Group 13 LN dissection, Group 14a LN dissection, Group 14v LN dissection, Group 15 LN dissection, Group 16a LN dissection, Group 16b LN dissection, combined organ resection, combined liver lobectomy, combined cholecystectomy, Combined splenectomy, combined resection of the body and tail of the pancreas and splenectomy, combined Whipple operation, combined oophorectomy, combined transverse colectomy, combined ascending and descending colectomy, remnant gastrectomy, jejunostomy, Broun anastomosis, infiltration of peripheral organs, infiltration of omentum, Liver infiltration, gallbladder infiltration, transverse mesocolon infiltration, transverse colon infiltration, pancreas head infiltration, pancreas body and tail infiltration, spleen infiltration, esophagus infiltration, duodenum infiltration, distant organ metastasis, liver metastasis, peritoneal metastasis, ovarian metastasis, extensive Lymph node metastasis, extensive abdominal metastasis, ascites, preoperative coronary heart disease, severe arrhythmia, hypertension, chronic obstructive pulmonary disease, chronic renal insufficiency, liver cirrhosis, portal hypertension, cerebrovascular disease, diabetes, weight loss, pyloric obstruction , Nutritional support before operation, nutritional support after operation, etc.; (4) Nominal variables (need to be dummyized and transformed into binary variables): digestive tract reconstruction method, gastrectomy range, tumor location, tissue type;

2. Firstly, the 79 variable indicators investigated were subjected to univariate analysis, and the corresponding statistical processing methods were as follows: continuous variables were tested by independent sample T test; ordered variables were tested by non-parametric tests (Mann-Whitney U test or Kolmogorov-Smirnov Z test); dichotomous variables adopt chi-square test or Fisher's exact probability method; the confidence interval (CI) is 95%, and the significant difference is P≤0.05. As a result, 18 variables with statistically significant differences were screened out as follows: Combined pancreas tail and splenectomy, combined Whipple operation, Borrman classification, intraoperative blood loss, age, tumor diameter, intraoperative time, liver function Child-Pugh score, preoperative chronic coexisting diseases, portal hypertension, total gastrectomy , Pyloric obstruction, Nol6a group lymph node dissection, No13 group lymph node dissection, tumor TNM staging, degree of lymph node dissection, remnant gastrectomy, prothrombin time;

3. Perform binary multiple unconditional logistic regression analysis on the 18 variables screened out, carry out model testing, discriminant analysis, calculate the partial regression coefficient and relative risk of each factor: OR=Exp(B), and obtain 8 true The factors that affect the serious complications after gastric cancer surgery are as follows: combined pancreas body and tail and splenectomy, degree of lymph node dissection, Child-Pugh score of liver function, chronic coexisting diseases before operation, total gastrectomy, lymph node dissection in No16a group, intraoperative blood loss, Tumor TNM staging;

4. Count the actual occurrence and predicted probability of postoperative complications for each patient, take the predicted probability as the test variable, and take the actual occurrence of postoperative complications as the state variable to perform receiver operating characteristic curve (ROC) analysis. The area under the curve (Az) evaluates the value of the prediction method, determines the best prediction cut-off value according to the Youden index, and evaluates the sensitivity and specificity of the prediction method;

5. According to the logistic regression analysis results of the above step 3, establish a prediction model for the risk of severe complications after gastric cancer surgery: P=Exp∑B0+B1X1+...+BkXk/1+Exp∑B0+B1X1+...+BkXk, where P is Response variable, representing risk probability value, X is independent variable, representing each risk factor, B is partial regression coefficient, combined with the optimal cut-off value determined by Youden index, it can be used to predict severe postoperative complications in each patient with gastric cancer risk probability.

The beneficial effects of the present invention are as follows:

In medical practice, the occurrence of a certain disease is often the result of the combined action of many pathogenic factors, and the causal relationship is intricate. When the Logistic Regression (Logistic Regression) fitting method is used for multivariate analysis, it can balance the effects of multiple confounding factors in a complex relationship, and the screened factors are more objective and credible, especially suitable for dependent variables that are dichotomous variables, The independent variable is the clinical data of multiple risk factors, and the analysis method has few restrictions on the distribution of the data, so it is particularly convenient for clinical use.

The receiver operating characteristic curve (Receiver Operating Characteristic Curve, ROC) analysis method is a curve obtained by using the sensitivity and (1-specificity) at different cut-off values as the vertical and horizontal coordinates respectively. The area of can be used to evaluate the prediction ability of the prediction system, and according to the Youden index, the best prediction cut-off value with high sensitivity and specificity can be determined.

The variable indicators of this prediction model are all highly objective clinical laboratory, imaging, surgery, and pathology indicators, so the reliability is extremely strong; the prediction model is confirmed by the Goodness of fit test and ROC analysis It has high accuracy, sensitivity and specificity.

The predictive model quantitatively analyzes the risk of serious complications after gastric cancer surgery, so as to help patients make medical decisions, assist ward management, and guide medical clinical research; at the same time, this set of surgical risk assessment methods and model establishment methods , and can also be widely used in operations other than gastric cancer.

Help patients make medical decisions: Evidence-based medicine (Evidence-based medicine, EBM) is the core model of modern clinical medical practice and research. One of the three basic principles of evidence-based medicine is to regard patients as the main participants in medical practice and one of the decision makers. The implementation of any diagnosis and treatment decision by a doctor must obtain the understanding and acceptance of the patient, and must take into account the patient's expectation and economic affordability of the treatment method. In clinical practice, the operation for some patients is doomed to have little effect, and due to the occurrence of serious postoperative complications, the quality of life is extremely poor, and the final death is still inevitable, which not only aggravates the pain of the patient, but also causes serious of medical waste. This prediction model can help patients participate in the medical decision-making of whether to perform surgery by objectively predicting the prognosis of surgery.

Auxiliary ward management: (1) Determining the patient's postoperative care level This prediction model has a unique role in determining whether the patient is admitted to the intensive care unit (ICU). According to statistics, among the current domestic ICU patients, low-risk patients who actually do not need to enter the ICU account for about 30%. By identifying low-risk and high-risk postoperative patients, this predictive model can help determine whether a patient needs intensive care, as well as the number of nurses required for each shift in the ward and the level of care required for each patient. In an era when medical expenses, especially intensive care expenses, are becoming more and more expensive, this prediction system can significantly reduce the economic burden of patients and save medical resources; (2) In the current medical situation, doctor-patient disputes have surged, partly because doctors The risk of surgery lacks an objective, quantifiable, and highly accurate prediction method, which makes patients have high expectations for the effect of surgery; through the surgical risk prediction system, it can help patients correctly understand the risks of surgery, help alleviate the contradiction between doctors and patients, and reduce the risk of surgery. Medical disputes; (3) Evaluate the level of treatment among different medical institutions and different treatment groups, analyze the utilization efficiency of limited resources, and even determine the annual health budget.

To guide medical clinical research and to evaluate new medical methods: for example, we select patients with a predicted postoperative mortality rate of 50-60% and divide them into two groups, one group is treated according to traditional methods, and the other group is treated with Early postoperative enteral nutrition support (EN), it was found that the actual mortality rate of the latter group was reduced to 30%, which indicated that EN can significantly reduce postoperative mortality and is worth popularizing.

4. Description of drawings

Figure 1 is the receiver operating characteristic curve (ROC curve) of this prediction method.

5. Specific implementation methods:

Embodiment 1: A method for predicting the risk of serious complications after gastric cancer surgery, the method steps are as follows:

1. Establishment of gastric cancer information database:

1.1 Sources of clinical data:

The subjects of the study were 1542 patients who underwent gastric cancer surgery in Jiangsu Provincial People's Hospital, Drum Tower Hospital of Nanjing University, and General Hospital of Nanjing Military Region from June 2002 to June 2006. A retrospective case-control study was adopted. The case group consisted of patients with severe complications after surgery, and the control group consisted of gastric cancer surgery patients without serious complications who were hospitalized at the same time.

1.2 Survey content All clinical data and information are subject to the original medical records, using uniform variable indicators, and all are input into the gastric cancer information database established with the SPSS13.0 statistical software package. A total of 79 indicators were selected from physical examination information, laboratory examination information, imaging examination information, surgical information, and tumor pathology information as possible risk factors for analysis, including: (1) continuous variables: number of lymph node metastases, lymph node dissection Number, intraoperative blood transfusion volume, intraoperative time, tumor diameter, age, plasma albumin, prealbumin, liver function Child-pugh score, total bilirubin, hemoglobin, white blood cell count, lymphocyte count, prothrombin time, Blood glucose, carcinoembryonic antigen; (2) Ordinal variables: degree of lymph node (LN) dissection, degree of radical surgery, T stage, N stage, TNM stage, degree of tumor differentiation, Borrman classification; the grading standards of ordinal variables are shown in Table 1 (3) Dichotomous variables: Group 10 LN dissection, Group 11p LN dissection, Group 12 LN dissection, Group 13 LN dissection, Group 14a LN dissection, Group 14v LN dissection, Group 15 LN dissection, Group 16a LN dissection, Group 16b LN dissection, combined organ resection, combined hepatectomy, combined cholecystectomy, combined splenectomy, combined resection of the body and tail of pancreas and splenectomy, combined Whipple operation, combined oophorectomy, combined transverse colectomy, Combined ascending and descending colectomy, gastrectomy, jejunostomy, Broun anastomosis, invasion of peripheral organs, invasion of omentum, invasion of liver, invasion of gallbladder, invasion of transverse mesocolon, invasion of transverse colon, invasion of pancreatic head, invasion of pancreatic body and tail, invasion Spleen, infiltration of esophagus, infiltration of duodenum, distant organ metastasis, liver metastasis, peritoneal metastasis, ovarian metastasis, extensive lymph node metastasis, extensive abdominal cavity metastasis, ascites, preoperative coronary heart disease, severe arrhythmia, hypertension, chronic Pulmonary obstruction, chronic renal insufficiency, liver cirrhosis, portal hypertension, cerebrovascular disease, diabetes, weight loss, pyloric obstruction, preoperative nutritional support, postoperative nutritional support, etc.; (4) nominal variable: digestive tract reconstruction method , gastrectomy range, tumor location, tissue type, etc. For nominal variables, dummy processing is required to convert them into multiple binary variables. Multivariate correlation matrices were then examined for collinearity, and variable deletion was performed if there was evidence of severe collinearity.

Table 1 Grading standards for ordinal variables (part) variable (X) score Analysis standard variable (X) score Analysis standard Borrman classification of gastric cancer by sex gastrectomy and digestive tract reconstruction 01123451231234 Cardia side pylorus side whole stomach combined visceral remnant stomach Biroux-I Biroux-IIRoux-en-YBor-IBor-IIBor-IIIBorIV Liver function grade Tumor location TNM staging 123123123456 Child-AChild-BChild-C Distal Stomach Proximal Stomach Whole Stomach IAIBIIIIIIAIIIBIV

1.3 Diagnostic criteria and occurrence of severe postoperative complications:

Serious postoperative complications are defined as potentially life-threatening complications that occur within 30 days after surgery, including: postoperative hemorrhage requiring reoperation, pulmonary infection, intestinal fistula, heart failure, acute renal failure, etc.; surgical death Defined as death from any cause within 30 days postoperatively.

The incidence rate of serious complications after radical gastrectomy in this group was 17.6% (271/1542), and the frequency of occurrence was pleural infection and pleural effusion, abdominal infection, dynamic ileus, incision infection, pancreatic fistula, anastomotic leakage, incision Dehiscence, acute input loop obstruction, duodenal stump fistula, intra-abdominal hemorrhage, anastomotic stoma obstruction, multiple organ dysfunction (including heart, lung, liver, kidney and other organ dysfunction), stress ulcer, gastroparesis, Acute pancreatitis, abdominal lymphatic fistula, acute cholecystitis, etc., some patients may have a variety of serious complications, and the operative mortality rate is 1.4% (21/1542).

2. Single factor analysis:

The selected 79 variables were used for univariate analysis. Continuous variables were tested by independent sample T test, ordinal variables were tested by non-parametric tests (Mann-Whitney U test or Kolmogorov-Smirnov Z test), and dichotomous variables were tested by Chi-square test or Fisher The exact probability method, the confidence interval is 95%, and the significant difference is P≤0.05. The statistical results (see Table 2, Table 3, and Table 4, respectively) showed that 18 of the 79 factors analyzed were closely related to severe complications after radical gastrectomy, namely, combined pancreatectomy and splenectomy, combined Whipple operation, Borrman classification, intraoperative blood loss, age, tumor diameter, intraoperative time, liver function Child-Pugh score, preoperative chronic coexisting diseases, portal hypertension, total gastrectomy, pyloric obstruction, No16a group lymph node dissection, No13 group lymph node dissection, tumor TNM staging, degree of lymph node dissection, remnant gastrectomy, prothrombin time.

Table 2 The statistical results of continuous variables relevant factor No complication group Complication group P value Age (years) Tumor diameter (cm) Intraoperative blood loss (ml) Intraoperative time (hours) Child-pugh integrated prothrombin time (seconds) 57.87±13.1685.015±2.6005232.05±189.6503.316±0.78945.17±0.44511.684±1.7919 62.13±12.9126.184±3.4508395.33±270.3513.754±1.14806.52±0.89413.892±1.8415 0.0130.0180.0010.0250.0090.012

Table 3 Statistical results of dichotomous variables relevant factor No complication group Complication group P value Preoperative chronic comorbidities portal hypertension remnant gastrectomy Total gastrectomy combined with Whipple operation for pyloric obstruction No.16a group lymph node dissection No.13 group lymph node dissection combined with pancreas body and tail and splenectomy (-)(+)(-)(+)(-)(+)(-)(+)(-)(+)(-)(+)(-)(+)(-)(+)(- )(+) 99527612353612323911061651250211170101995276976295121952 159112240312432819081253182323913313814512623833 0.0000.0200.0350.0140.0090.0310.0040.0020.000

Table 4 Nonparametric test results of graded data relevant factor Wilcoxon W Z value P value Lymph node dissection (0-D0, 1-D1, 2-D2, 3-D3, 4-D4) TNM stage (1-IA, 2-IB, 3-II, 4-IIIA, 5-IIIB, 6-IV ) Borrman classification of advanced gastric cancer (1-Bor I, 2-BorII, 3-BorIII, 4-BorIV) 26597.000103559.500103410.500 -1.337-2.633-3.073 0.0020.0280.016

3. Binary multiple unconditional logistic regression analysis

The 18 variables that were preliminarily screened out were subjected to binary multiple unconditional logistic regression analysis (regression method), model testing and discriminant analysis were performed to calculate the partial regression coefficient and relative risk of each factor: OR=Exp(B). The results showed that a total of 8 factors were entered into the logistic regression model, and the order of effect was: combined pancreas body and tail and splenectomy (OR=3.422), degree of lymph node dissection (OR=2.967), liver function Child-Pugh score (OR= 2.012), preoperative chronic comorbidity (OR=1.961), total gastrectomy (OR=1.501), 16a group lymph node dissection (OR=1.391), intraoperative blood loss (OR=1.207), tumor TNM staging (OR=1.119 ) (Table 8). The model test results (Table 5, 6) show that: the regression equation has significant significance; the discriminant test results (Table 7) show that: the model has a high prediction accuracy (85.1%).

Table 5 Model Test 1 (Fitness Test) Chi-square df Sig. Step 1 StepBlockModel 357.948357.948357.948 343434 .000.000.000

Table 6 Model Test 2 -2Log Likelihood Cox & Snell R Square Nagelkerke R Square Step 1 205.037 .572 .776

Table 7 Discriminant test table Actual cases Model prediction example total Accuracy(%) + - + 209 62 271 - 168 1103 1271 total 377 1165 1542 85.1

Table 8 Binary multiple unconditional logistic regression analysis results independent variable (X) Regression coefficient (B) Standard Deviation (SE) Wald value (Wald) df P value (Sig) Relative risk (OR) 95.0% CI for OR Lower Unner 1. Combined pancreas tail and splenectomy 2. Lymph node dissection 3. Liver function Child-Pugh score 4. Preoperative chronic coexisting diseases 5. Total gastrectomy 6. No16a group lymph node dissection 7. Intraoperative blood loss 8. Tumor TNM Staging 9. Constant 1.2300.0210.0091.0410.8920.8040.0030.016-2.942 0.7410.0150.0100.4260.4700.5630.0970.1800.327 2.7531.1070.0182.4353.1391.9430.0120.0091.357 111111111 0.0120.0150.0290.0340.0560.1030.1490.2880.043 3.4222.9672.0121.9611.5011.3911.2071.1190.813 0.8000.9860.9890.9070.9150.7270.7180.649 14.63714.01213.7627.9745.7856.6036.7625.230

4. Receiver operating characteristic curve (ROC) analysis

The actual occurrence and predicted probability of postoperative complications of each patient were counted, with the predicted probability as the test variable and the actual occurrence of postoperative complications as the state variable, the receiver operating characteristic curve was analyzed to obtain the area under the curve (Az ); Youden (Youden) index=sensitivity+specificity-1, with the maximum point of Youden index judged as the best cut-off value of prediction probability, the results are shown in Figure 1 and Table 9:

The area under the ROC curve (Az) is 83.3%, indicating that the prediction model has a good predictive value (when Az is 0.5-0.6, it means that the predictive value is low, when Az is 0.6-0.8, it means that the predictive value is medium, and when Az is 0.8-1.0 , indicating high predictive value); the critical value (P) corresponding to the maximum point of Youden’s index is 0.391 (that is, when P<0.391, it is judged that no serious complications will occur after surgery; when P>0.391, it is judged that postoperative Serious complications will occur), at this time, the predicted sensitivity is 85.7%, and the specificity is 78.6%.

Table 9 Determining the cut-off value of the best prediction probability based on Youden index predicted probability value sensitivity 1-specificity Youden index 0.0000.0190.027......0.3020.3380.391*0.4450.479......0.9130.9761.000 1.0001.0001.000......0.9090.8790.857*0.8030.673......0.1660.0150.000 1.0000.9710.941......0.5590.3910.214*0.2050.195......0.0290.0000.000 0.0000.0290.059......0.3500.4880.643*0.5980.478......0.1370.0150.000

Further analysis of the relationship between the predicted probability grade and the degree of postoperative severe complications (see Table 10), the results confirmed that: the higher the predicted probability (P value) grade, the more serious the degree of postoperative severe complications, that is, the higher the risk (P<0.001 ).

Table 10 The relationship between the probability level of the model and the severity of postoperative complications P value severity of postoperative complications total contingency coefficient (1) (2) (3) (4) low risk medium risk high risk 0.4～0.60.6～0.80.8～1.0 31196 233825 183129 41631 7610491 0.434P<0.001

5. Establish a prediction model for the risk of severe complications after gastric cancer surgery

According to the binary multiple unconditional logistic regression analysis results in the above step 3, the prediction model is established as follows: P(1)=Exp∑(-2.942+1.23X1+0.021X2+0.009X3+1.041X4+0.892X5+0.804X6+ 0.003X7+0.016X8)/[1+Exp∑(-2.942+1.23X1+0.021X2+0.009X3+1.041X4+0.892X5+0.804X6+0.003X7+0.016X8)], set the prediction probability cut-off value to 0.391 ; When P<0.391, it is judged that there will be no serious complications after operation, and when P>0.391, it is judged that there will be serious complications after operation. The larger the P value, the greater the possibility of serious complications.

Below by specific case the present invention will be further described:

The relevant information of a patient with gastric cancer is as follows: According to the condition, he (she) needs to perform radical total gastrectomy (X5=1), combined with resection of the body and tail of the pancreas and splenectomy (X1=1), and the degree of lymph node dissection is D3 (X2 =3), lymph nodes around the abdominal aorta were dissected (that is, lymph nodes in No.16a group need to be dissected, X6=1), the Child-Pugh score of the preoperative liver function was 5 points (X3=5), and the pathological stage of the tumor was IIIA ( X8=4), if there is no chronic coexisting disease before operation (X4=0), the blood loss during operation is expected to be 200ml (X7=200), and it is substituted into the prediction equation to get P=0.680. Since P>0.391, it is predicted that the patient Severe postoperative complications may occur, 0.6<P<0.8, indicating that the risk of severe complications is moderate.

Claims (1)

1. A method for predicting the risk of severe complications after gastric cancer surgery, characterized in that the method uses binary multiple unconditional logistic regression analysis technology and receiver operating characteristic curve analysis technology to evaluate the serious complications after gastric cancer surgery The degree of risk, the prediction method includes the following steps: (1) The SPSS 13.0 software package was used to establish a gastric cancer information database, and the recorded 79 variables were as follows: (1) Continuous variables: number of lymph node metastasis, number of lymph node dissection, intraoperative blood transfusion, intraoperative time, tumor diameter, age , plasma albumin, prealbumin, liver function Child-pugh score, total bilirubin, hemoglobin, white blood cell count, lymphocyte count, prothrombin time, blood sugar, carcinoembryonic antigen; (2) ordered variable: lymph node ( LN) degree of dissection, surgical radical degree, T stage, N stage, TNM stage, degree of tumor differentiation, Borrman classification; (3) dichotomous variables: group 10 LN dissection, group 11p LN dissection, group 12 LN dissection , Group 13 LN dissection, Group 14a LN dissection, Group 14v LN dissection, Group 15 LN dissection, Group 16a LN dissection, Group 16b LN dissection, combined organ resection, combined liver lobectomy, combined cholecystectomy , Combined splenectomy, Combined pancreas tail and splenectomy, Combined Whipple operation, Combined oophorectomy, Combined transverse colectomy, Combined ascending and ascending colectomy, Remnant gastrectomy, Jejunostomy, Broun anastomosis, Peripheral organ infiltration, Omentum infiltration , liver infiltration, gallbladder infiltration, transverse mesocolon infiltration, transverse colon infiltration, pancreas head infiltration, pancreas body and tail infiltration, spleen infiltration, esophagus infiltration, duodenum infiltration, distant organ metastasis, liver metastasis, peritoneal metastasis, ovarian metastasis, Extensive lymph node metastasis, extensive abdominal metastasis, ascites, preoperative coronary heart disease, severe arrhythmia, hypertension, chronic obstructive pulmonary disease, chronic renal insufficiency, liver cirrhosis, portal hypertension, cerebrovascular disease, diabetes, weight loss, pylorus Obstruction, preoperative nutritional support, postoperative nutritional support, etc.; (4) Nominal variables, which need to be dummyed and transformed into binary variables: digestive tract reconstruction method, gastrectomy extent, tumor location, tissue type; (2) First, the 79 variable indicators investigated were subjected to univariate analysis, and the corresponding statistical processing methods were as follows: continuous variables used independent sample T-test; ordinal variables used non-parametric tests, namely Mann-Whitney U test or Kolmogorov -Smirnov Z test; Chi-square test or Fisher's exact probability method was used for dichotomous variables; the confidence interval (CI) was 95%, and the significant difference was P≤0.05. As a result, 18 variables with statistically significant differences were selected as follows : Combined pancreas body and tail and splenectomy, combined Whipple operation, Borrman classification, intraoperative blood loss, age, tumor diameter, intraoperative time, liver function Child-Pugh score, preoperative chronic coexisting diseases, portal hypertension, whole stomach Resection, pyloric obstruction, No16a group lymph node dissection, No13 group lymph node dissection, tumor TNM staging, degree of lymph node dissection, remnant gastrectomy, prothrombin time; (3) Perform binary multiple unconditional logistic regression analysis on the 18 variables screened out, carry out model testing, discriminant analysis, calculate the partial regression coefficient and relative risk of each factor: OR=Exp(B), and draw 8 The factors that really affect the serious complications after gastric cancer surgery are as follows: combined pancreas body and tail and splenectomy, degree of lymph node dissection, Child-Pugh score of liver function, chronic coexisting diseases before operation, total gastrectomy, lymph node dissection in No16a group, intraoperative blood loss , Tumor TNM staging; (4) Statistically analyze the actual occurrence and predicted probability of postoperative complications for each patient, take the predicted probability as the test variable, and take the actual occurrence of postoperative complications as the state variable to analyze the receiver operating characteristic curve. The area evaluates the value of the prediction method, determines the best prediction cut-off value according to the Youden index, and evaluates the sensitivity and specificity of the prediction method; (5) According to the logistic regression analysis results of the above step 3, a prediction model for the risk of severe complications after gastric cancer surgery is established: P=Exp∑B0+B1X1+...+BkXk/1+Exp∑B0+B1X1+...+BkXk, where P is the dependent variable, representing the risk probability value, X is the independent variable, representing each risk factor, B is the partial regression coefficient, combined with the optimal cut-off value determined by the Youden index, it can be used to predict the occurrence of severe postoperative complications in each patient with gastric cancer disease risk probability.

Images