Clinical and Procedural Predictive Model for Pneumothorax Risk After CT-Guided Lung Biopsy

To develop and validate a model integrating clinical/imaging and procedural variables to predict pneumothorax after CT-guided percutaneous lung nodule biopsy.

We retrospectively enrolled 395 patients and split them 7:3 into training (n=276) and validation (n=119) cohorts. Variables with P < 0.15 in group comparisons (combined with clinical relevance) were selected as candidates and analyzed using age- and sex-adjusted univariable logistic regression. Two prespecified logistic models were fitted: a baseline model (clinical/imaging variables only) and a full model (baseline plus procedural variables). Discrimination, calibration, and clinical utility were assessed using ROC curves, calibration plots, and decision curve analysis (DCA).

Pneumothorax occurred in 22.1% (training) and 22.7% (validation). The baseline model contained BMI and emphysema grade; the full model additionally included patient position (prone or lateral vs supine) and the number of needle adjustments. In training, the full model outperformed the baseline (AUC 0.722 vs 0.630; P=0.018). In validation, AUC was 0.720 vs 0.670 (P=0.307), but the difference did not reach statistical significance. Sensitivity was higher with the full model (0.667 vs 0.593). Both models showed good calibration; the full model was closer to the ideal line across predicted probabilities of 0.1-0.5. DCA indicated greater net benefit for the full model across most threshold probabilities.

A model combining clinical/imaging and procedural characteristics may facilitate peri-operative risk communication and support peri-procedural risk management for CT-guided lung nodule biopsy.