A predictive model to distinguish malignant and benign thyroid nodules based on age, gender and ultrasonographic features / Modelo preditivo para diferenciação entre nódulos malignos e benignos da tireoide com base na idade, sexo e características ultrassonográficas

Abstract Introduction: A discussion in literature about a standardized decision support tool for the management of thyroid nodules remains. Objective: The purpose of this study was to create a statistical prediction model for thyroid nodules management. Methods: Two hundred and four benign and 57 malignant thyroid nodules were selected for a retrospective study. The variables age, gender and ultrasonographic features were examined using univariate and multivariate models. A statistical formula was used to calculate the risk of cancer of each case. Results: In multivariate analysis, irregular shape, absence of halo, lower mean age, homogeneous echotexture, microcalcifications and solid content were associated with cancer. After applying the formula, 20 cases (7.6%) with a calculated risk for malignancy ≤3.0% were found, all of them benign. Setting the calculated risk in ≥80%, 21 (8.0%) cases were selected, and in 85.7% of them cancer was confirmed in histopathology. Internal accuracy of the prediction formula was 92.5%. Conclusions: The prediction formula reached high accuracy and may be an alternative to other decision support tools for thyroid nodule management.

Resumo Introdução: Persiste na literatura uma discussão sobre uma ferramenta padronizada de apoio à decisão para o manejo de nódulos tireoidianos. Objetivo: Criar um modelo de previsão estatística para o manejo de nódulos tireoidianos. Método: Foram selecionados 204 casos de nódulos tireoidianos benignos e 57 malignos para o estudo retrospectivo. As variáveis idade, sexo e características ultrassonográficas foram analisadas com modelos univariados e multivariados. Uma fórmula estatística foi usada para calcular o risco de câncer de cada caso. Resultados: Na análise multivariada, a forma irregular, a ausência de halo, menor idadesubp média, ecotextura homogênea, microcalcificações e conteúdo sólido foram associadas ao câncer. Após a aplicação da fórmula, foram encontrados 20 casos (7,6%) com risco calculado de malignidade ≤ 3,0%, todos benignos. Definiu-se o risco calculado em ≥ 80%, 21 casos (8,0%) foram selecionados e em 85,7% deles o câncer foi confirmado pela histopatologia. A precisão interna da fórmula de previsão foi de 92,5%. Conclusões: A fórmula de previsão alcançou alta precisão e pode ser uma opção para outras ferramentas de apoio à decisão para o manejo de nódulos da tireoide.

A predictive model to distinguish malignant and benign thyroid nodules based on age, gender and ultrasonographic features.

INTRODUCTION: A discussion in literature about a standardized decision support tool for the management of thyroid nodules remains. OBJECTIVE: The purpose of this study was to create a statistical prediction model for thyroid nodules management. METHODS: Two hundred and four benign and 57 malignant thyroid nodules were selected for a retrospective study. The variables age, gender and ultrasonographic features were examined using univariate and multivariate models. A statistical formula was used to calculate the risk of cancer of each case. RESULTS: In multivariate analysis, irregular shape, absence of halo, lower mean age, homogeneous echotexture, microcalcifications and solid content were associated with cancer. After applying the formula, 20 cases (7.6%) with a calculated risk for malignancy ≤3.0% were found, all of them benign. Setting the calculated risk in ≥80%, 21 (8.0%) cases were selected, and in 85.7% of them cancer was confirmed in histopathology. Internal accuracy of the prediction formula was 92.5%. CONCLUSIONS: The prediction formula reached high accuracy and may be an alternative to other decision support tools for thyroid nodule management.