Two Predictive Models Of Central Compartment Lymph Node Metastasis In Thyroid Micropapillary Carcinoma

Background and purposePapillary thyroid microcarcinoma(PTMC)refers to papillary thyroid carcinoma(PTC)with a diameter of ? 1 cm.The incidence of thyroid cancer has continued to rise in countries around the world in the past few decades.More than 50% of patients with new thyroid cancer are micropapillary carcinoma,and the mortality rate is still quite stable.Although the degree of malignancy of PTMC is generally low,the prognosis is good after standardized treatment,but its CLNM rate is no less than PTC.In 2010,thyroid cancer was the fourth most common tumor type among women in some European countries and the United States.In Korea,thyroid cancer ranks first among all malignant tumors in the female population,and the Ministry of Health statistics show that thyroid cancer is rising.To the third place in female malignant tumors.Lin's systematic review based on 707 articles concluded that neck ultrasound and subsequent cytology of fine needle aspiration can identify thyroid cancer early,but it is unclear whether population-based or targeted screening can Reducing mortality or improving vital patient outcomes,screening for the identification of indolent thyroid cancer,and treatment of these over-diagnosed cancers may increase the risk of patient harm.In addition,exposure to diagnostic radiation and obesity may also be one of the reasons.At the same time,the rate of thyroid micropapillary carcinoma(PTMC)surgery has increased.According to predictions,it is estimated that by 2018,thyroid cancer will become the most common cancer among women,and this increased economic burden will reach US$3,107,000,000.The huge cost has been widely discussed as an increasingly serious clinical and public health problem,but there is still much controversy from the diagnosis and treatment of micropapillary cancer.In an era when network applications are becoming more and more common,we generate a lot of data every day.The traditional data analysis system has become increasingly weak.Many new methods of data analysis have emerged.They are not only efficient,but also can be connected with today's information technology.Decision tree models and nomograms are two products of the era of big data analysis.This article explores the risk factors for lymph node metastasis in the central region of thyroid micropapillary carcinoma.Method:PTMC patients who were first admitted to the General Hospital of Tianjin Medical University from January 2016 to December 2018 were selected as subjects.Inclusion criteria 1)initial thyroid cancer surgery;2)postoperative pathology confirmed as thyroid micropapillary carcinoma(PTMC);3)thyroid gland + isthmus resection / total resection plus central lymph node dissection;4)complete clinical And pathological data.Exclusion criteria 1)recurrent thyroid cancer;2)no incidental thyroid cancer without central lymph node dissection.Univariate analysis was performed by chi-square test,single factor logistic regression analysis,and multivariate analysis was performed by unconditional logistic regression analysis.P<0.05 was considered statistically significant.The decision tree model modeling process was performed using SPSS Modeler 14.1(SPSS,USA)statistical software.In all the included samples,90% of the samples were automatically selected by the software as training samples for machine learning and decision tree modeling.The C5.0 algorithm is applied in the decision tree construction process and analyzed by Pearson's chi-square test to determine whether the node is segmented.After establishing the decision tree,the remaining 10% samples are used as verification samples.The accuracy of the model is tested by using the software's own analysis node,and the sensitivity and specificity are calculated.The test level ? value is doubled.The process of creating a nomogram uses R 386 3.5.2 software,using the "rms","regplot" software package to construct the normal nomogram and interactive nomogram,and pass the area under the curve of the ROC curve.AUC)compares and predicts the degree of agreement between the CLNM rate and the actual CLNM rate to assess and quantify the accuracy of the nomogram to predict the outcome of the outcome.Result:Chi-square test and Logistic single factor analysis were performed on each factor.The results showed gender(OR=0.44,95% CI=(0.25-0.75)P<0,05),age(OR=0.45,95% CI=(0.28-0.73)P<0,05),the number of primary tumors(OR=1.71,95% CI=(1.03-2.84)P<0,05)was significant,and the maximum diameter of the tumor,the presence or absence of calcification under B-ultrasound,B The number of super-tumor,tumor invasion and capsule,TSH level,main complaint time,B-ultrasound aspect ratio,Bultrasound tumor edge characteristics,tumor unilateral or bilateral,thyroid antibody level,Tg level,merge bridge There was no significant difference in the relationship between thyroiditis and CLNM(P>0.05).Logistic multivariate regression model analysis showed that women(OR=0.37,95% CI=(0.21-0.65)P<0,05),age<45 years(OR=0.39,95% CI=(0.24-0.64)P <0,05),tumor diameter>0.8cm(OR=1.70,95% CI=(1.02-2.84)P<0,05),primary foci(OR=2.15,95% CI=(,1.24-3.73)P<0,05)was an independent influencing factor of CLNM(P<0.05),and there was no significant difference in whether there was calcification under B-ultrasound and whether the tumor invaded the capsule(P>0.05).As shown by the nomogram: male,age <45 years old,tumor size,multiple primary tumors,tumor invasion of the thyroid capsule and central lymph node metastasis of thyroid micropapillary carcinoma have a strong correlation,and whether calcification under ultrasound The tumor does not invade the capsule and the tumor invades the capsule,and the maximum diameter correlation of the tumor under ultrasound is relatively weak.The internal alignment curve is plotted and the internal calibration curve is drawn(Fig.3).It can be seen that the initial C-index of the nomogram is the average absolute probability of the actual risk probability and the predicted risk probability after 1000 internal verifications.The error is 0.032,and the ROC curve is established by the probability predicted by the nomogram(Fig.4).The area under the curve is AUC=0.695,95% CI=(0.636-0.755),P<0.05.In the decision tree model,gender,age,age 30 and 49,multiple primary tumors,tumor invasion of the thyroid capsule can assist in the determination of lymph node metastasis in the central region of thyroid micropapillary carcinoma.The sensitivity of the training sample was 75.4%,the specificity was 69.8%,the accuracy was 71.26%,and the Kappa coefficient was 0.379,(P<0.05).The sensitivity of the validated sample was 70.0%,the specificity was 53.8%,and the accuracy was 58.33%.The Kappa coefficient was 0.187(P>0.05).Conclusion:1.Logistic multivariate regression analysis,nomogram,and decision tree prediction models were found in males,ages <45 years old,tumor size,multiple primary tumors,tumor invasion of thyroid capsule and thyroid micropapillary carcinoma.Lymph node metastasis has a strong correlation.2.Logistic multivariate regression analysis,nomogram,and decision tree prediction model have different advantages and characteristics in preoperative prediction of central lymph node metastasis of thyroid micropapillary carcinoma.Logistic multivariate regression analysis is the most traditional The method is convenient to understand;the nomogram model visualizes the effect size of various factors,which can make the role of the factors at a glance;the decision tree prediction model provides a tree structure,for the segmentation of factors,the selection of critical points Very meaningful.3.Our study did not analyze data from long-term follow-up,such as disease recurrence and disease-free survival;therefore,we cannot directly conclude that the scoring system can predict prognosis.Second,the study was conducted in a crosssectional design rather than a longitudinal view.Third,single-center verification is not accurate enough.The scoring system requires multi-center validation to check its predictive value.Further investigation is required for a longer period of time.