Purpose <p>Pulmonary contusion is a common injury in thoracic trauma patients and a significant risk factor for respiratory complications such as pneumonia. This study aimed to investigate the role of pulmonary contusion volume and radiomics as predictors of pneumonia within 7 days of trauma and the degree of overlap between contused lung segments and those where pneumonia subsequently developed.</p> Methods <p>This retrospective observational study included all consecutive adult trauma patients admitted to the ED with a whole-body CT scan available within 6&#xa0;h from admission. The primary outcome was the occurrence of early-onset pneumonia (EOP). Data were extracted from clinical records and CT scan images were analysed for volumetric and radiomic analysis. ROC curves were built, and robust Poisson regression and extreme gradient boosting models were developed to evaluate the most influential clinical and radiologic variables.</p> Results <p>Among 486 patients included, 79 (16.3%) developed early-onset pneumonia. Volumetric analysis showed higher contused lung volumes and proportions in pneumonia patients. The contusion percentage had the highest AUC (0.644) for predicting pneumonia over T-AIS and AAST lung injury scales. Multivariable analysis identified H-AIS, contusion proportion, and P/F ratio as significant predictors, with a model AUC of 0.710. Radiomic analysis improved prediction accuracy, with an extreme gradient boosting model showing an estimated out-of-sample AUC of 0.790.</p> Conclusion <p>Pulmonary contusion volume is an interesting predictor of early-onset pneumonia in thoracic trauma patients. Advanced radiology techniques, including radiomics, could enhance prediction accuracy and aid in early risk stratification, potentially improving patient outcomes.</p> Trial registration <p>NCT06750822, retrospectively registered.</p>

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Pulmonary contusion volume and radiomics as predictors of early-onset pneumonia in trauma patients, a retrospective single-centre study

  • Lorenzo Gamberini,
  • Giovanna Maria Astuti,
  • Bianca Biavati,
  • Davide Boarato,
  • Matteo Tontodonati,
  • Irene Palmitessa,
  • Matteo Laruccia,
  • Carlo Coniglio,
  • Alessio Bertini,
  • Paolo Emilio Orlandi,
  • Marco Tartaglione,
  • Davide Allegri,
  • Antonio Bruno

摘要

Purpose

Pulmonary contusion is a common injury in thoracic trauma patients and a significant risk factor for respiratory complications such as pneumonia. This study aimed to investigate the role of pulmonary contusion volume and radiomics as predictors of pneumonia within 7 days of trauma and the degree of overlap between contused lung segments and those where pneumonia subsequently developed.

Methods

This retrospective observational study included all consecutive adult trauma patients admitted to the ED with a whole-body CT scan available within 6 h from admission. The primary outcome was the occurrence of early-onset pneumonia (EOP). Data were extracted from clinical records and CT scan images were analysed for volumetric and radiomic analysis. ROC curves were built, and robust Poisson regression and extreme gradient boosting models were developed to evaluate the most influential clinical and radiologic variables.

Results

Among 486 patients included, 79 (16.3%) developed early-onset pneumonia. Volumetric analysis showed higher contused lung volumes and proportions in pneumonia patients. The contusion percentage had the highest AUC (0.644) for predicting pneumonia over T-AIS and AAST lung injury scales. Multivariable analysis identified H-AIS, contusion proportion, and P/F ratio as significant predictors, with a model AUC of 0.710. Radiomic analysis improved prediction accuracy, with an extreme gradient boosting model showing an estimated out-of-sample AUC of 0.790.

Conclusion

Pulmonary contusion volume is an interesting predictor of early-onset pneumonia in thoracic trauma patients. Advanced radiology techniques, including radiomics, could enhance prediction accuracy and aid in early risk stratification, potentially improving patient outcomes.

Trial registration

NCT06750822, retrospectively registered.