Increasing diagnostic accuracy for thyroid nodules by an integrated multivariate approach: a methodological study.

Objectives: Current guidelines for diagnostic management of thyroid nodules are based on a linear approach, using categorial classifications to cluster diagnostic findings and they still lead to unnecessary surgery. A diagnostic scoring system, based on clinical, cytological and ultrasound findings is described. Materials and Methods: Two groups of patients (168 and 55 pts) were used to compute a multivariate model and the discriminating threshold by ROC curves. The performance of the derived scoring system was assessed by a simulation on a third group of 60 patients, who had undergone surgery according to current guidelines. Results: The scoring system displayed a sensitivity of 100%, specificity 53.3%, positive and negative predictive values of 68.1% and 100%. According to the scoring system, 16 out of 60 operations would have been saved. Conclusions: A scoring system can take into account in a more accurate way the full informative content of the fine-grained description of diagnostic and clinical features.