Is the Parsimonious Eurolung risk model a good predictor for all populations? An external validation study

INTRODUCTION: Over 20 models have been developed in the last 30 years to predict postoperative mortality after thoracic surgery. The Parsimonious Eurolung score, introduced by Brunelli et al. in 2019, is a simplified logistic regression model for predicting perioperative morbidity (EuroLung1) and mortality (EuroLung2) based on the European Society of Thoracic Surgeons (ESTS) database. While validated for European populations, no studies have confirmed its applicability in other populations with different backgrounds. OBJECTIVES: To externally validate the Parsimonious Eurolung risk models using a Portuguese cohort of patients undergoing lung cancer resection. METHODS: This retrospective analysis included patients with non-small cell lung cancer (NSCLC) who underwent resection surgery at a Portuguese centre between January 2018 and December 2021. The Parsimonious Eurolung 1 (morbidity) and Parsimonious Eurolung 2 (mortality) models were used to predict morbidity and 30-day mortality. Model validation assessed calibration, discrimination, and clinical usefulness. RESULTS: A total of 410 patients were included. 4 (1.25%) patients died within 30 days. Mortality did not differ between types of surgery. 78 patients experienced significant morbidity, with pneumonectomies associated with highest morbidity rates. The mean Parsimonious Eurolung 1 score (morbidity) was 13.49% (SD 6.84) and the mean Parsimonious Eurolung 2 score (mortality) was 2.17% (SD 2.13). Morbidity scores differed significantly across years (p = 0.011) and were significantly higher in patients with complications (p < 0.001). For Parsimonious Eurolung 1, an AUC of 0.668 was calculated, with the calibration plot showing an underestimation for mid-range values and significant overfitting. For Parsimonious Eurolung 2, an AUC of 0.437 was obtained, showing gross underestimation of values. CONCLUSIONS: External validation was partially achieved. The Parsimonious Eurolung 1 score is moderately applicable for morbidity. The Parsimonious Eurolung 2 score is not validated for 30-day mortality in this cohort. It should not guide surgical eligibility, and this point can be generalized to other low-mortality settings. This study provides valuable insights into the performance of the Parsimonious Eurolung models in a specific national context while highlighting the ongoing challenges in risk prediction for thoracic surgery. The findings suggest a need for quality multicentric databases and more granular, population-specific risk models, potentially incorporating machine learning approaches.