| BackgroundEsophageal cancer(EC)is known to the most common malignant in the world,the International Agency for Research on cancer(IARC)updated the authoritative statistics in September 2018.Esophageal cancer is the seventh morbidity among all the malignancy which is one place lower than before,while the mortality rate remains the sixth,showing its danger.For China,which has the largest population base,esophageal cancer has the highest morbidity and mortality in the world.In clinical practice,the high mortality rate is mainly caused by the high incidence of early postoperative complications of esophageal cancer,especially on Anastomosis Leakage after Esophagectomy.Although great progress has been made in the treatment of esophageal surgery of peri-operation in recent decades,the prediction and prevention of ALE by clinicians mostly remain in the state of empirical prediction,which is not conducive to the evaluation of postoperative recovery status of patients,and may even delay the diagnosis and treatment of early complications.Due to previous clinical workers are limited by clinical conditions,to be thoughtless about the influence factors of Anastomosis Leakage after Esophagectomy,the research cases failed to layered,making the results of the study barely satisfactory,but the emergence of artificial neural network model,broke the situation,which use its forecast analysis model that had been set up,can not only considering the interaction between each influence factor,but also through the mass of self-learning can achieve self-denial of the relevance between influence factor and disease,matches the prediction efficiency of the best path eventually,compared with the traditional Logistic regression analysis,there is no clear result on which method is better in predicting the effect of complications,which needs to be proved by theoretical data urgently.ObjectiveBuilding up the model of Artificial Neural Network and Logistic Regression Analysis respectively,to determine the high correlation factors and to find the suited forecasting model for Anastomosis Leakage after Esophagectomy,the two methods of analysis of the prediction model,predict performance are compared,the analysis methods and predictive performance of these two predictive models are summarized and compared,and the advantages and disadvantages of the two models are clarified,which will lead to the prediction of postoperative complications of other clinical diseases in the future.MethodsUsing the method of retrospective study on 538 cases of patients who were received and cured by the group of director of thoracic surgeons in our hospital during April 2010 to December 2018 in our hospital,to analyze the data in the clinical information that before surgery and intraoperative,such as gender,age,body mass index(BMI),smoking history,whether there is related chronic diseases(hypoalbuminemia,diabetes,chronic obstructive pulmonary disease,cardiovascular disease),clinical classification and staging of esophageal cancer,minimally invasive surgery or not,operation time and blood loss,whether Preoperative neoadjuvant chemoradiotherapy and postoperative prophylactic tracheotomy were used and so on influence factor and the occurrence of Anastomosis Leakage after Esophagectomy.Results(1)A total of 538 patients with esophageal cancer who underwent surgical treatment in the same treatment group of thoracic and cardiac surgery in our hospital were collected,in which that patient with Anastomosis Leakage after Esophagectomy of esophageal has 60 patients with a positive rate of 11.2%.(2)The chi-square test was used for univariate analysis of the factors(hypoproteinemia,location of cancer,etc.)that may affect postoperative complications of esophageal cancer obtained from literature research,which accorded with P<0.25.In the inspection method,there were nine factors that can be included as prediction variable,including of sex(P = 0.028),hypoalbuminemia(P<0.001),diabetes(P = 0.140),cardiovascular disease(P = 0.066),adenocarcinoma(P = 0.225),squamous cell carcinomas(P = 0.225),MIE(P = 0.114),preoperative neoadjuvant chemotherapy(P = 0.058)and postoperative prophylactic tracheostomy(P = 0.123),among those prediction variable hypoalbuminemia(P<0.001)has a significant effect on complication of esophageal anastomotic fistula;In addition,univariate analysis of continuous data by T test showed that age(P=0.962)and BMI(P=0.391)were excluded,Rank-sum test of TNM staging showed that T(P=0.317),N(P=0.550)and M(P=0.221)were all excluded except M staging can intake in predictive variables.(3)Further using stepwise regression method to analyze various factors,we got the best efficiency of logistic regression prediction model(the x^2 of H-L fitting test= 1.617,df = 1,P = 0.204)which was the most significant predicting variables take effect as hypoalbuminemia,preoperative neoadjuvant chemotherapy and postoperative prophylactic tracheostomy,the accuracy can reach 91.8%,at this time the AOR values of those three were 38.784,1.869 and 1.275 respectively,the 95%confidence interval was 16.233-87.823,0.484-7.216 and 0.159-10.246 respectively.At this point,the persistence sample was substituted into the model again,and the accuracy,sensitivity and specificity were obtained as follows:93.44%,64.70%and 96.38%.(4)The three-layer model was established with the automatic architecture of artificial neural network.There were 20 units in the input layer,6 units in the hidden layer,and 2 units in the output layer,taking 0.5 as the cutting point,which can obtain the accuracy,sensitivity and specificity of persistence sample accuracy were 94.53%,70.59%and 96.99%,respectively.(5)ROC curves were drawn respectively,and we obtained that AUC of Logistic regression was 0.881,and AUC of artificial neural network was 0.845.Therefore,the efficacy of the two models in predicting postoperative complications of esophageal cancer was basically the same,however Artificial Neural Network had slight advantages over Logistic Regression Analysis Mode in predicting Anastomosis Leakage after Esophagectomy.ConclusionWe can draw the following two conclusions through this research,for the first one is that gender,hypoalbuminemia,diabetes mellitus,cardiovascular disease,adenocarcinoma,squamous carcinoma,surgical operation,preoperative neoadjuvant chemotherapy,postoperative prophylactic tracheostomy and the transfer of M can be used as predictor variable in complication of Anastomosis Leakage after Esophagectomy,therein hypoalbuminemia is the independent risk factors for complication of Anastomosis Leakage;Secondly,people think of Artificial Neural Network in the prediction of view holds an enormous advantage questionable in the past,remain to be discussed,in terms of medical forecast,the accuracy between Artificial Neural Network prediction model with Logistic regression Analysis models are much the same,but the Artificial Neural Network has the ability of autonomic learning and convenient data processing,in the case of effectiveness is not falling behind can replace the Logistic regression Analysis model. |
PDF Full Text Request