Development of a risk prediction model of pneumothorax in percutaneous computed tomography guided transthoracic needle lung biopsy

INTRODUCTION To retrospectively evaluate the incidence of and the risk factors for pneumothorax and intercostal catheter insertion (ICC) after CT-guided lung biopsy and to generate a risk prediction model for developing a pneumothorax and requiring an ICC. METHODS 255 CT-guided lung biopsies performed for 249 lesions in 249 patients from August 2014 to August 2019 were retrospectively analysed using multivariate logistic regression analysis. Risk prediction models were established using backward stepwise variable selection and likelihood ratio tests and were internally validated using split-sample methods. RESULTS The overall incidence of pneumothorax was 30.2% (77/255). ICC insertion was required for 8.32% (21/255) of all procedures. The significant independent risk factors for pneumothorax were lesions not in contact with pleura (P < 0.001), a shorter skin-to-pleura distance (P = 0.01), the needle crossing a fissure (P = 0.004) and emphysema (P = 0.01); those for ICC insertion for pneumothorax were a needle through emphysema (P < 0.001) and lesions in the upper lobe (P = 0.017). AUC of the predictive models for pneumothorax and ICC insertion were 0.800 (95% CI: 0.745-0.856) and 0.859 (95% CI: 0.779-0.939) respectively. Upon internal validation, AUC of the testing sets of pneumothorax and ICC insertion were 0.769 and 0.822 on average respectively. CONCLUSION The complication rates of pneumothorax and ICC insertion after CT-guided lung biopsy at our institution are comparable to results from previously reported studies. This study provides highly accurate risk prediction models of pneumothorax and ICC insertion for patients undergoing CT-guided lung biopsies.