Construction of the prognostic model for assessing the risk of medullary thyroid cancer metastases using binary logistic regression

Abstract

Background. Medullary thyroid cancer is a topical disease that is often accompanied by metastases. The danger of this pathology requires timely and adequate surgery. Objective to assess the prognostic value and informativeness of some clinical indicators with the selection of the most optimal and reliable potential factors in the development of the mathematical equation for calculating the personal probability of detecting metastases of medullary thyroid cancer in the absence of clinical and instrumental signs in the preoperative stage. Materials and methods. Patients with medullary thyroid carcinoma with and without locoregional metastases participated in this study. To verify metastatic deposits, a pathomorphological study was performed using the TNM classification (UICC). StatPlus Pro v.7.6, EpiTools and MedCalc statistical calculators were used for statistical processing of results. Data Mining technologies were used to assess the degree of impact of potential predictors using the data mining add-on for MS Office. To assess the diagnostic value of the test, ROC analysis was performed and the corresponding characteristic curve was constructed with the calculation of the area under it (according to the DeLong method). For the operational characteristics of the tests, 95 % confidence interval was calculated according to the Wilson method. The results were considered statistically significant at p < 0.05. Results. Basal calcitonin, patient sex, multifocality, and total tumor size have been shown to be significant in the a priori of the medullary thyroid cancer metastatic risk assessment. These indicators can be used not only as predictors of unfavourable prognosis, but also as indicators for individual determination of the surgery scope. Conclusions. The method of binary logistic regression to assess latent metastasis showed lower sensitivity (0.77 vs 0.89) and higher specificity (0.90 vs 0.64) in contrast to the monofactorial prognosis based on preoperative calcitonin levels. Further improvement of the model requires additional analysis of erroneous test results in the applied training sample. The proposed prognostic model due to calculations in MS Excel allows you to easily and quickly obtain information, so it can be used as an additional diagnostic tool when choosing a method of surgical treatment.