Development of a logistic regression formula for evaluation of subcentimeter thyroid nodules

The purpose of this study was to build a logistic regression formula for ameliorating the diagnosis of subcentimeter thyroid nodules.

The sonographic features of 889 subcentimeter nodules were reviewed retrospectively with reference of histologic results. The diagnostic performance of each feature was evaluated. Multivariate binary logistic regression was used to develop the formula for evaluation of subcentimeter nodules, and the cutoff value was decided for recommending biopsy.

The logistic regression formula was -0.029age - 2.063US1 - 0.812US2 + 1.781US3 + 1.627height-to-width ratio - 0.333 (for nonhypoechogenicity, US1 = 1; for hypoechogenicity, US1 = 0; for a well-circumscribed margin, US2 = 1 and US3 = 0; for a microlobulated margin, US2 = 0 and US3 = 1; and for an irregular margin, US2 = 0 and US3 = 0). The area under the receiver operating characteristic curve for the formula was 0.860. The sensitivity, specificity, positive predictive value, negative predictive value, and accuracy of the formula under the cutoff value of 0.284 were 90.9%, 54.0%, 66.4%, 85.6%, and 72.5%, respectively. With the use of the formula, 229 of 830 nodules could avoid surgery.

A logistic regression formula with a cutoff value could provide an objective and easy tool with effective diagnostic performance, which could improve diagnosis of subcentimeter thyroid nodules and reduce unnecessary biopsy, decreasing costs and patient discomfort.