Predictors of pulmonary metastases on chest computed tomography in children and adolescents with osteosarcoma—tips for qualifying patients for thoracotomy

Abstract

BackgroundOsteosarcoma is the most common primary malignant bone tumour in children and adolescents. Lungs are the most frequent and often the only site of metastatic disease. The presence of pulmonary metastases is a significant unfavourable prognostic factor. Thoracotomy is strongly recommended in these patients, while computed tomography (CT) remains the gold imaging standard. The purpose of our study was to create tools for the CT-based qualification for thoracotomy in osteosarcoma patients in order to reduce the rate of useless thoracotomies.MethodsSixty-four osteosarcoma paediatric patients suspected of lung metastases on CT and their first-time thoracotomies (n = 100) were included in this retrospective analysis. All CT scans were analysed using a compartmental evaluation method based on the number and size of nodules. Calcification and location of lung lesions were also analysed. Inter-observer reliability between two experienced radiologists was assessed. The CT findings were then correlated with the histopathological results of thoracotomies. Various multivariate predictive models (logistic regression, classification tree and random forest) were built and predictors of lung metastases were identified.ResultsAll applied models proved that calcified nodules on the preoperative CT scan best predict the presence of pulmonary metastases. The rating of the operated lung on the preoperative CT scan, dependent on the number and size of nodules, and the total number of nodules on this scan were also found to be important predictors. All three models achieved a relatively high sensitivity (72–92%), positive predictive value (81–90%) and accuracy (74–79%). The positive predictive value of each model was higher than of the qualification for thoracotomy performed at the time of treatment. Inter-observer reliability was at least substantial for qualitative variables and excellent for quantitative variables.ConclusionsThe multivariate models built and tested in our study may be useful in the qualification of osteosarcoma patients for metastasectomy through thoracotomy and may contribute to reducing the rate of unnecessary invasive procedures in the future.