Development and validation of a risk nomogram predicting pneumothorax requiring chest tube placement post-percutaneous CT-guided lung biopsy

Masha Bondarenko, Jianxiang Zhang, Ulysis Hugo Baal, Brian Lam, Gunvant R. Chaudhari, Yoo Jin Lee, Jamie Lee Schroeder, Maya Vella, Brian Haas, Thienkhai Vu, Kimberly Kallianos, Jonathan Liu, Shravan Sridhar, Brett Elicker, Jae Ho SohnBMC Medical Imaging 2025 · DOI

Introduction

CT-guided transthoracic lung biopsy is a standard, minimally invasive procedure for diagnosing pulmonary nodules. However, it carries a significant 20–30% risk of pneumothorax. In approximately 2–17% of procedures, the pneumothorax is severe enough to require the surgical placement of a chest tube, which increases hospital stays, healthcare costs, and infection risks. While clinicians recognize factors like age and emphysema as contributors to this risk, existing predictive models often rely on subjective radiologist annotations or complex "black box" algorithms that are impractical for bedside use. This study develops and externally validates a transparent, web-based risk nomogram that leverages automated quantitative emphysema assessment to predict high-risk patients.

Methods and Study Design

The researchers conducted a retrospective two-center study to develop a clinically usable predictive tool. To objectively measure lung disease, they adapted an automated quantitative emphysema algorithm. This algorithm analyzes biopsy CT scans using 3x3x3 voxel kernels and Hounsfield Unit (HU) thresholding to classify emphysema severity into a 3-point scale (None, Mild/Moderate, Severe), removing the variability inherent in subjective text reports.

For model development, the authors prioritized simplicity. They utilized Akaike Information Criterion (AIC) forward variable selection to identify the most impactful covariates. The final logistic regression model was translated into a visual nomogram and an interactive web-based risk calculator.

Datasets and Cohort Selection

The study utilized a total of 2,512 patient records across two distinct hospital environments:

  • Training Dataset (UCSF): 1,899 patients from the University of California, San Francisco, a tertiary academic referral center (1994–2022).
  • Validation Dataset (ZSFG): 613 patients from Zuckerberg San Francisco General Hospital, a county safety-net hospital (2014–2023).

Across the combined cohort, 157 patients (6.7%) required a chest tube following a biopsy-induced pneumothorax.

Flowchart detailing patient inclusion, exclusion criteria, and the model development pipeline.

Key Results and Model Performance

The final model identified three primary predictors of chest tube placement:

  1. Non-prone patient position (Odds Ratio: 2.48)
  2. Severe emphysema (Odds Ratio: 1.99)
  3. Age over 60 (Odds Ratio: 1.80)

The model demonstrated robust generalizability. It achieved a Harrell’s C-index (AUC) of 0.664 on the internal UCSF cohort and improved to 0.715 on the external ZSFG validation cohort. The nomogram also showed excellent calibration, with a mean absolute error of only 0.5% in the internal cohort.

ROC curves illustrating the predictive performance (AUC) of the model on both the internal UCSF and external ZSFG datasets.

Discussion and Clinical Implications

The study highlights that simplicity drives clinical utility. By distilling the risk profile into three accessible factors, the authors created a tool that clinicians can easily use at the bedside. The success of the quantitative emphysema algorithm proves that objective computer vision techniques can be effectively tied to clinical outcomes, outperforming subjective human interpretation. Furthermore, the finding that non-prone positioning is a major risk factor suggests that procedural adjustments, such as favoring prone positioning to stabilize the lung, could potentially reduce complications.

Limitations

The study is limited by its retrospective nature and the inherent subjectivity in the decision to place a chest tube, which varies by institutional protocol. The modest AUC values suggest that additional unmeasured variables, likely related to granular procedural details not captured in standard reports, influence pneumothorax risk. Future work should involve prospective validation and the integration of more advanced imaging features to further refine predictive accuracy.

Summary updated

BibTeX
@article{bondarenko2025development,
  author = {Masha Bondarenko and Jianxiang Zhang and Ulysis Hugo Baal and Brian Lam and Gunvant R. Chaudhari and Yoo Jin Lee and Jamie Lee Schroeder and Maya Vella and Brian Haas and Thienkhai Vu and Kimberly Kallianos and Jonathan Liu and Shravan Sridhar and Brett Elicker and Jae Ho Sohn},
  title = {{Development and validation of a risk nomogram predicting pneumothorax requiring chest tube placement post-percutaneous CT-guided lung biopsy}},
  journal = {BMC Medical Imaging},
  year = {2025},
  doi = {10.1186/s12880-025-01794-y},
}
Figure