| Study 1 Construction of a prognostic model for patients with pancreatic cancer and suspected peripancreatic vein invasionBackground:Pancreatic cancer has a high degree of malignancy and its prognosis is extremely poor.Therefore,usually at the time of diagnosis,the disease is often in advanced stages.However,whether or not pancreatic cancer invades the peripancreatic vessels and the extent of the invasion often affects the choice of treatment strategies.By reviewing previous studies,we found that there is currently no predictive model for predicting survival prognosis in patients with pancreatic cancer surrounding the pancreas.Therefore,the purpose of this study was to establish a prognostic evaluation model that relied on preoperative imaging.Method:The clinical data of 239 patients with pancreatic cancer with suspected peripancreatic vein involvement who underwent surgical treatment were collected in four large hospitals in China from January 2012 to December 2016.The related hospitals for data collection included the Second Hospital Affiliated to Medical College of Zhejiang University,the Oriental Hepatobiliary Surgery Hospital Affiliated to Second Military Medical University,Changhai Hospital Affiliated to Second Military Medical University and Tongji Hospital Affiliated to Huazhong University of Science and Technology.Patient data from three hospitals were selected as the modeling group(178 patients)and the patients in the other hospital were used as the external validation group.Because the study was about preoperative predictive models,only demographic variables and imaging variables were included for univariate and multivariate analysis.According to the results of multi-factor analysis,the prediction model was constructed by using R language,and the quality of the model was evaluated by using C-index and correction curve.Finally,the model was validated by external verification group.In addition,we also analyzed the correlation between the degree of deformation of the venous image and the pathological manifestations of the peripancreas by using the chi-square test.Result:According to multivariate analysis,when the age was(P=0.001),length of contact between tumors and veins was(P=0.035),degree of vein deformation was(P<0.001)and involvement of lymph nodes was(P=0.001),these were all independent risk factors affecting the prognosis of these patients.According to the patient’s prognosis,we divided the six manifestations of imaging vascular abnormalities(non-contact of tumor blood vessels,mild vascular deformation,teardrop sign,cluster sign,stenosis>1/2,complete occlusion of blood vessels)into 4 categories,including Category 1(tumor vessel non-contact),Category 2(vascular dysplasia/teardrop sign),Category 3(clustered sign or stenosis>1/2),and Category 4(complete vascular occlusion).These four categories of patients showed different prognostic manifestations(Category 1 vs.Category 2:P=0.024;Category 2 vs.Category 3:P=0.004;and Category 3 vs.Category 4:P<0.001).The established prognostic prediction model had good predictive ability.At the same time,the C-index values of the model group and the external validation group reached 0.814 and 0.824,respectively.However,there was also a high correlation between the degree of venous deformation around the pancreas and pathological results(model group:P<0.0001;external validation group:P=0.001).Conclusion:In this study,the model for predicting the prognosis of patients with pancreatic cancer with suspected percutaneous invasion of the pancreas had achieved the ideal predictive effect.Study 2 The construction of a model for predicting the invasion of the peripancreatic vein wall in patients with pancreatic cancerBackground:Pancreatic cancer is a malignant tumor of the digestive tract that is very malignant and difficult to diagnose and treat.In recent years,the morbidity and mortality of pancreatic cancer have increased significantly,and the disease is usually in advanced stage at the time of diagnosis.The main problem that has plagued the majority of surgeons has been the assessment of pancreatic cancer with peripancreatic venous invasion.A wrong assessment can lead to a wrong treatment direction and delay the patient’s condition.For example,unnecessary vascular resection and reconstruction in patients with non-invasive veins may lead to increased postoperative complications and mortality,thus affecting the prognosis of patients.On the other hand,with the promotion of preoperative neoadjuvant chemotherapy,apply neoadjuvant chemotherapy for patients who have not been subjected to the same vein may lead to tumor resistance or even progression.Therefore,preoperative assessment of venous invasion of the pancreas is particularly important.By reviewing previous studies,we found that although several models have been reported for predicting the invasion of peripheral pancreatic veins,these models have some limitations.The purpose of this study was to construct a more accurate predictive model of the invasion of peripancreatic vein.Method:The clinical data of pancreatic cancer patients with suspected peripancreatic vein involvement who underwent surgery in the Second Affiliated Hospital of Medical College of Zhejiang University from January 2012 to January 2017 were collected.At the same time,combined with the clinical data collected in "Paper 1" from January 2012 to December 2016,at the Eastern Hepatobiliary Surgery Hospital affiliated to the Second Military Medical University,Changhai Hospital affiliated to the Second Military Medical University,and Tongji Hospital affiliated to Huazhong University of Science and Technology.A total of 247 patients were collected,and 181 patients were randomly selected as model group,and the remaining 66 patients were selected as validation group,so that the ratio of the two groups was maintained at about 3:1.From the aspects of demographics,biology,clinical pathology,patients themselves and anatomy,the independent risk factors affecting the invasion of blood vessels around the pancreas were calculated by univariate and multivariate analysis.According to the results of multi-factor analysis,the prediction model was constructed by using R language,and the quality of the model was evaluated by using C-index and correction curve.Finally,the model was validated by external verification group.The scoring results of the model would be included in the risk grouping of patients and recommended for treatment.Result:According to the analysis of multiple factors,the length of contact between the tumor and the vein(P=0.031),the degree of deformation of the vein(P=0.001),and the degree of tumor surrounding the vein(P=0.048)were the independent risk factors for vascular invasion that affected around the pancreas of the patient.The established prognostic prediction model had good predictive ability.At the same time,the C-index values of the model group and the external validation group reached 0.896 and 0.963,respectively.According to the score,patients were divided into three groups:low-risk group(the possibility of venous invasion<50%),median group(the possibility of venous invasion<50%-90%)and high-risk group(the possibility of venous invasion>90%).In addition,we found that the prognosis of this type of patient was only related to whether the vein was invaded,and not related to the depth of invasion of the vein wall.Conclusion:Peripancreatic veins were invaded by pancreatic cancer,mainly by structural factors.In this study,the prediction model had achieved the desired results.Study 3 The construction of a prognostic model of patients with pancreatic cancer and suspected peripancreatic vein invasion based on CE-CT characteristicsBackground:Once pancreatic cancer patients have contracted the peripancreatic veins,this period is usually in the late stages of pancreatic cancer.In previous studies,we constructed an imaging-based prognostic assessment model to predict the prognosis of pancreatic cancer patients with suspected peripancreatic vein invasion.Although the model had achieved good prediction results(the C-index values of the model group and the external verification group reached 0.814 and 0.824,respectively),with the continuous development of image omics,we hope to upgrade the predicting model we built earlier to further improve its predecting capabilities.The purpose of this study was to construct a prognostic model combining imaging features with clinical factors.Method:A total of 74 patients with pancreatic cancer with suspected peripancreatic vein involvement underwent surgery in the Second Affiliated Hospital of Medical College of Zhejiang University were collected and selected as model group.In addition,25 cases of patients in the Eastern Hepatobiliary Surgery Hospital affiliated to the Second Military Medical University were collected and selected as external validation group.A total of 396 imaging features were extracted from pre-processed CE-CT images in each patient.By using LASSO regression method,the best image omics features were screened,and based on this,an imaging omics prediction model was established.Subsequently,based on the scores of the imaging ensemble model combined with clinical factors,independent risk factors for image prognosis were found by univariate and multivariate analysis.The prediction model was constructed by using R language,and the quality of the model was evaluated by using C-index and correction curve.Finally,the model was validated by external verification group.Result:In this study,the obtained imaging histology was highly correlated with the prognosis of patients with pancreatic cancer and peripancreatic vein invasion(model group:P=0.013;external validation group:P=0.018).According to the multivariate analysis,the imaging omics score(P=0.000),CA199(P=0.004),the degree of venous defonnation(P=0.000),and the involvement of lymph nodes(P=0.001)were the independent risk factors that affected the prognosis of these patients.The established prognostic prediction model had good predictive ability.At the same time,the C-index values of the model group and the external validation group reached 0.833 and 0.836,respectively.Conclusion:In this study,the accuracy of the model established for predicting the prognosis of patients with pancreatic cancer and suspected peripancreatic vain invasion was further improved. |
PDF Full Text Request