Development and validation of a nomogram for predicting chest drainage after CT-guided percutaneous lung biopsy: a...

To develop and externally validate a ridge logistic regression model containing only pre-procedural predictors to estimate the individual probability of requiring chest drainage after CT-guided percutaneous core-needle lung biopsy.

This multicenter retrospective study included 2,492 patients. All 2,092 patients who underwent biopsy at Institution 1 from 2015 to 2025 were used for model development, and 400 patients treated at Institution 2 in 2025 formed the external-validation cohort. Nine variables available before needle entry were prespecified: age, sex, planned position, emphysema, lesion short-axis diameter, longitudinal lesion location, anatomical lobe, planned fissure traversal, and visceral-pleural adjacency. A ridge-penalized logistic model was fitted; the penalty was selected by stratified 10-fold cross-validation using the one-standard-error rule, and the complete modeling process was internally validated with 1,000 bootstrap resamples. The frozen model was applied directly to Institution 2 without refitting, recalibration, or threshold optimization. Development and external-validation sample sizes were formally assessed retrospectively using the Riley frameworks.

Chest drainage was required in 127/2,092 patients (6.1%) at Institution 1 and 34/400 (8.5%) at Institution 2. At Institution 1, the apparent AUC was 0.786 (95% CI, 0.745-0.827) and the optimism-corrected AUC was 0.766. For the frozen model at Institution 2, the AUC was 0.683 (95% CI, 0.579-0.779) and the Brier score was 0.0755 (95% CI, 0.0550-0.0982). The external calibration intercept was 0.41 (95% CI, 0.00-0.72), calibration slope was 1.13 (95% CI, 0.52-1.85), and observed-to-expected ratio was 1.44 (95% CI, 1.00-1.89), indicating systematic underprediction. The development sample was slightly smaller than required under conservative assumptions. Under the prespecified stringent precision targets, external validation would require approximately 17,397 patients and 1,479 events, whereas the available cohort contained 400 patients and 34 events. External validation should therefore be interpreted as independent but preliminary.

The pre-procedural model showed moderate discrimination in an independent external institutional cohort but underestimated absolute risk. Prospective multicenter validation, with local recalibration where appropriate, is required before routine clinical use.