Construction And Validation Of Predictive Model For Preoperative Diagnosis Of Thyroid Carcinoma

Objective: The preoperative diagnosis of thyroid cancer has always been a challenge.This study aims to establish three thyroid cancer risk prediction models using cost-effective and easily available demographic characteristics,ultrasound signs and clinical biomarkers:(1)Build a risk prediction model for thyroid nodules with malignant and benign nodules to improve the prediction efficiency of thyroid nodules.(2)To explore the potential risk factors of CLNM in PTC patients,and to construct and validate the predictive model of CLN diagnosis.(3)PTMC nomogram model was constructed to predict the possibility of CLNM in PTMC patients.Methods: The patients who went to the Affiliated Hospital of Traditional Chinese Medicine of Xinjiang Medical University from 2018 to 2021 for thyroid ultrasound examination and were finally diagnosed as thyroid nodules by pathology after surgery were selected retrospectively as the subjects of this study.(1)In this study,3877 patients with thyroid nodules were included in the diagnosis and analysis of thyroid malignant and benign nodules.They were divided into training set and verification set according to the ratio of7:3.The LASSO regression model was used for factor selection to build a risk prediction model for thyroid nodule diagnosis.(2)In this study,1820 patients with PTC diagnosed pathologically after surgery were included.R(4.1.1)was used for data analysis.Logistic regression was used to analyze the influencing factors of lymph node metastasis in the central region of PTC patients,and a risk prediction model was constructed.(3)A total of2036 PTMC patients were randomly divided into two groups,the training group(n=1527)and the validation group(n=509).LASSO regression model was used for factor selection,and multivariable logistic regression analysis was conducted to check the risk factors related to CLNM and establish a nomograph for predicting CLNM.Finally,the above three studies all used the area under the ROC curve(AUC),calibration curve and decision curve analysis(DCA)to evaluate the discrimination ability,accuracy and clinical practicability of the model.Results:(1)Candidate variables were selected through LASSO regression and multivariate logistic regression analysis.Age,echo,morphology and color Doppler flow imaging(CFDI)were identified as independent risk factors for identifying benign thyroid nodules.The area under the ROC curve was 0.847(95%CI: 0.830-0.864),the corresponding cut-off value was 0.694,the specificity was 0.742,and the sensitivity was 0.820;TI-RADS classified 1915 benign nodules and 1962 malignant nodules.The specificity,sensitivity and AUC values were 82.2%,66.0% and 0.752(95%CI: 0.736-0.768),respectively.The area under ROC curve of ultrasound imaging combined with TI-RADS was 0.867(95%CI:0.852-0.883),and the specificity and sensitivity were 0.733 and 0.864 respectively.DCA results show that the model’s net income is better when the threshold is in the range of 0.7-0.9.(2)The results showed that gender,age,total diameter of cancer focus,extraglandular invasion,cervical lymph node enlargement,total thyroid hormone(TT4)and antithyroid peroxidase(TPO)were independent risk factors for lymph node metastasis in central area of PTC patients(P<0.05).By establishing a prediction model,the area under the ROC curve(AUC)value of lymph node metastasis in the central area of PTC in the training set was0.732(95%CI:0.705-0.759),and the AUC value in the validation set was 0.731(95%CI:0.690-0.773).Hosmer-Lemeshow goodness of fit test showed a good degree of fit(P>0.05);The analysis of decision curve shows that when the threshold probability of patients is 0～0.6,the nomograph prediction model predicts that the net income of risk is higher.(3)LASSO regression model shows that 22 variables(non-zero coefficients)may be the influencing factors of CLNM.Subsequently,multivariate logistic regression analysis showed that younger age,male,calcification,1 ≥ diameter ≥ 0.5,multifocal,ETE,ECLN,lower Na and higher Tg Ab were the final risk factors of CLNM(P<0.05).By establishing the prediction model,the area under the ROC curve(AUC)of CLNM in the training group and the verification group were 0.734(95%CI: 0.703-0.765)and 0.740(95%CI: 0.683-0.797),respectively.The model has good calibration ability,and the average absolute errors of the training group and the verification group are 0.006 and 0.022 respectively.The DCA results showed that the model was useful in clinical practice when the intervention was determined in the training group and the validation group within the threshold probability range of 0.09-0.72 and 0.11-0.77,respectively.Conclusion: The three nomogram models constructed in this study respectively described the accurate identification of thyroid malignant and benign nodules,the prediction of lymph node metastasis in the central region of PTC patients,and the preoperative diagnosis of CLNM in PTMC patients.The nomograph models have good predictive effect,and they are expected to become an effective tool to indicate the progress and predict the overall survival of thyroid cancer patients in clinical practice,and provide support for clinical workers to formulate personalized treatment plans for patients.