PneumoScore: Risk Prediction Model for 90-Day Mortality After Lung Resection.

BACKGROUND

Lung resection is the gold-standard treatment for early stage lung cancer, but remains associated with significant mortality, highlighting the need for accurate preoperative risk prediction. This study developed a novel model to predict 90-day mortality after lung cancer resection and compared its performance with well-established models.

METHODS

Data were extracted from the Brazilian Lung Cancer Registry, a multicentric database including 2885 patients who underwent lung resection between 2002 and 2024. Records missing more than 15% of variables or any essential variable were excluded. The dataset was randomly divided into training (70%) and testing (30%) subsets. Candidate predictors were selected on the basis of clinical relevance and univariable analyses and multiple machine learning algorithms were evaluated. Final performance was assessed by area under receiver operating characteristic curve (AUROC), area under precision-recall curve (AUPR), Brier Score, Brier Skill Score (BSS), calibration slope, and intercept, using nonparametric bootstrapping to quantify variability.

RESULTS

In total, 2001 patients were included, of whom 88 (4.40%) died within 90 days post surgery. The best-performing model, PneumoScore, was a logistic regression incorporating age, sex, predicted postoperative FEV1%, ASA ≥ IV, coronary artery disease, cerebrovascular disease, congestive heart failure, pneumectomy, thoracotomy, and extended resection. PneumoScore achieved an AUROC of 0.84 (0.77-0.91), AUPR of 0.99 (0.98-1.00), Brier Score of 0.033 (0.022-0.045), BSS of 0.256 (0.180-0.332), intercept of 0.067 (-0.045, 0.179), and slope of 1.067 (0.912-1.222).

CONCLUSIONS

PneumoScore demonstrated excellent performance for predicting 90-day mortality, a critical yet underexplored outcome in thoracic surgery, supporting its potential to enhance preoperative evaluation of patients with lung cancer.