External validation of the eurolung risk model for morbidity using an institutional database on minimally invasive major lung resections

Abstract

Background

The EuroLung1 model was developed by the European Society of Thoracic Surgeons (ESTS) to estimate the risk of major cardiopulmonary complications after anatomical lung resection. Because minimally invasive approaches were underrepresented in the original derivation cohort, its performance in contemporary thoracoscopic practice remains uncertain. We therefore conducted an external validation using a cohort of patients undergoing minimally invasive major lung resection.

Methods

This retrospective study included 644 consecutive patients (293 females and 351 males; median age, 66 years) treated with thoracoscopic lobectomy or segmentectomy. For each patient, the EuroLung1 score was determined according to the published scoring algorithm incorporating eight weighted clinical and operative variables. Patients were subsequently categorized into the predefined risk classes. Associations between the score and major morbidity were examined, while model discrimination and calibration were evaluated using standard validation metrics.

Results

The cohort included 574 lobectomies (89.1%) and 70 segmentectomies (10.9%). Histopathology revealed 548 patients (85.1%) with confirmed non-small cell lung cancer (NSCLC), 66 (10.2%) with benign disease, and 30 (4.7%) with pulmonary metastasis. Postoperative mortality and major complication rates were 0.9% and 12.9%, respectively. In logistic regression analysis, each 1-point increase in the EuroLung1 score was associated with 9% higher odds of morbidity [odds ratio (OR) =1.092, 95% confidence interval (CI): 1.010–1.180, P=0.03]. Similarly, advancing by one risk class conferred a 33% increase in the odds (OR =1.327, 95% CI: 1.027–1.715, P=0.03). Calibration analysis revealed that the EuroLung1 score underestimated morbidity risk in lower-risk classes and overestimated it in higher-risk classes. Discrimination performance of EuroLung1 score was poor, with an area under the receiver operating characteristic (ROC) curve of 0.573.

Conclusions

Although the EuroLung1 score retained a significant relationship with postoperative major morbidity after minimally invasive anatomical lung resection, its predictive performance was suboptimal. The weak discrimination and variable calibration observed in this validation cohort suggest that model recalibration or revision may be necessary before application to individual risk estimation in contemporary minimally invasive thoracic surgery.

Highlight box.

Key findings

• In an external validation cohort of 644 patients undergoing minimally invasive major lung resection, the EuroLung1 score remained significantly associated with postoperative major cardiopulmonary morbidity.

• Discrimination was poor, with an area under the receiver operating characteristic curve of 0.573.

• Calibration varied across risk classes, with underestimation of morbidity in lower-risk patients and overestimation in higher-risk patients.

What is known and what is new?

• The EuroLung1 score was developed from the European Society of Thoracic Surgeons database to predict major postoperative morbidity after anatomical lung resection.

• Previous external validation studies suggested reduced performance in cohorts with a high proportion of minimally invasive procedures.

• This study provides validation in a cohort undergoing exclusively minimally invasive lobectomy or segmentectomy and demonstrates that statistical association with morbidity persists, but predictive performance remains limited.

What is the implication, and what should change now?

• The EuroLung1 score should be interpreted cautiously when estimating individual morbidity risk in contemporary minimally invasive thoracic surgery patients.

• Recalibration and model updating using modern minimally invasive datasets may improve predictive accuracy.

• Future risk prediction models should incorporate contemporary patient populations and clinically relevant predictors to enhance discrimination and calibration.

Introduction

In the last decade, minimally invasive anatomic lung resections have been widely adopted and are currently preferred for the treatment of early-stage non-small cell lung cancer (NSCLC) due to better preservation of pulmonary function and less trauma to the chest wall compared to thoracotomy, particularly in high-risk patients (1,2). However, the positive effect of minimally invasive approach on the postoperative outcome are not properly reflected in the current guidelines for the preoperative risk assessment or physiologic evaluation (3,4). In contrast, surgical approach (minimally invasive versus thoracotomy) was explicitly incorporated as a key predictor in the EuroLung risk models published in 2017 and updated in 2020 (5,6). Developed by Brunelli et al. using 47,960 resections from the European Society of Thoracic Surgeons (ESTS) database, EuroLung1 score represents a morbidity model based on eight predictors. Each predictor is assigned a weight of 1 to 3 points and the summed score stratifies patients into six risk classes for cardiopulmonary complications.

Due to the small proportion of minimally invasive cases (13.1%) in the derivation cohort, concerns have arisen regarding the transportability of the EuroLung1 score to contemporary practice, where the minimally invasive approaches predominate (4,7). Indeed, external validations in cohorts with higher rates of minimally invasive surgery have shown limited generalizability of EuroLung1 score (8,9). To date, few studies have evaluated its performance in a cohort undergoing exclusively minimally invasive major lung resection (10). The ESTS Database Committee has explicitly called for further independent validations (11). These gaps bolstered our interest to conduct an external validation using the database on minimally invasive lobectomy or segmentectomy of the Klinikum Braunschweig. We present this article in accordance with the STROBE reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0821/rc).

Methods

A retrospective analysis was conducted using the database on thoracoscopic major lung resections of the Klinikum Braunschweig. The cohort comprised 644 consecutive patients who underwent intended thoracoscopic lobectomy or segmentectomy from 2009 through 2017. Before surgery, 606 patients had confirmed or suspected malignant disease (NSCLC or pulmonary metastasis), while 38 patients had a preoperatively established benign diagnosis. Final histopathological examination revealed NSCLC in 548 patients (85.1%), pulmonary metastasis in 30 patients (4.7%), and benign disease in 66 patients (10.2%). The composition of the study cohort and the transition from preoperative diagnosis to final histopathological diagnosis are summarized in Figure 1. Patient data were not part of the ESTS database. The study was approved by the Ethics Committee of the Hannover Medical School (ID: 019/2016BO2), and the requirement for individual informed consent was waived due to the retrospective nature. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Figure 1.

Patient flow diagram of the study cohort showing preoperative diagnostic categories and final histopathological diagnoses. NSCLC, non-small celllungcancer.

The EuroLung1 morbidity score was calculated for each patient in strict accordance with the original derivation by Brunelli et al. (5). Specifically, eight clinical predictors [male sex, age >65 years, predicted postoperative forced expiratory volume in 1 second (FEV₁) <70%, coronary artery disease, cerebrovascular disease, chronic kidney disease, thoracotomy and extended resection] were each assigned a weight of 1, 2, or 3 points. The total score, ranging from 0 to 19, was obtained by summing the individual point values. This aggregate score was then used to stratify patients into six predefined morbidity risk classes. The “thoracotomy” variable was defined based on the intended surgical approach (intention-to-treat principle), regardless of intraoperative conversion from minimally invasive approach to thoracotomy.

All patients with suspected or confirmed NSCLC underwent thorough staging workup including computed tomography, positron emission tomography and brain magnetic resonance imaging, as previously described (12). Cases with mediastinal nodal involvement on imaging were evaluated further by endobronchial ultrasound-guided transbronchial needle aspiration or cervical mediastinoscopy. Final pathologic staging followed the American Joint Committee on Cancer Cancer Staging Manual, 8th edition, and all cases were discussed in our institutional (Klinikum Braunschweig) multidisciplinary lung tumor board prior to surgery.

Physiologic evaluation was undertaken in all patients prior to surgery as described in the prior publication of our group (13). The predicted postoperative value of forced expiratory volume in the first second expressed as a percent (ppoFEV₁%) and of diffusing capacity of lung expressed as a percent (ppoDLCO%) were calculated using the functional segment technique in the present study. Patients with ppoFEV₁% or ppoDLCO% <40% underwent further risk stratification, either by stair-climb test or cardiopulmonary exercise testing (cycle ergometry), with a threshold of ≥10 mL/kg/min (or ≥35% predicted VO_2max) deemed acceptable for resection.

As described in our prior study, thoracoscopic lobectomy and segmentectomy were performed using a three-port approach, including a 3-cm anterolateral access incision in the fourth intercostal space without rib spreading and with visualization only through the monitor (14). Lobar vessels and the bronchus were individually divided. Hilar and mediastinal lymph nodes were dissected.

Patients were routinely followed at 6 weeks and 3 months after surgery in the outpatient clinic. Postoperative mortality and morbidity were defined as death and adverse events within 90 days of surgery according to The Society of Thoracic Surgeons (STS)/ESTS joint standardization of variable definitions and terminology, consistent with the definitions applied in the ESTS Database used for development of the original EuroLung1 model (15). Specifically, coronary artery disease was defined as a history of angina pectoris, myocardial infarction, or prior interventions like coronary artery bypass grafting or percutaneous coronary intervention. Cerebrovascular disease was defined as a history of clinically confirmed stroke or transient ischemic attack presumed to be of vascular origin. Chronic kidney disease was defined as the presence of kidney damage or an estimated glomerular filtration rate (eGFR) of less than 60 mL/min/1.73 m², persisting for 3 months or more. Thoracotomy was defined according to the intended surgical approach (intention-to-treat principle). Extended resection was defined as an anatomic lung resection combined with resection of chest wall, pancoast tumours, resection of the atrium or superior vena cava, diaphragm resection, vertebral resection, pleuropneumonectomy, sleeve pneumonectomies, intrapericardial pneumonectomy. Major morbidity was defined as any cardiopulmonary complication; overall morbidity included any other complications (e.g., bronchopleural fistula, empyema, delirium), but excluded purely technical events (e.g., recurrent laryngeal nerve injury, chylothorax), consistent with our prior methodology (16).

Statistical analysis

Categorical variables are presented as counts and percentages; continuous variables as median with interquartile range (IQR). Group comparisons were performed using Fisher’s exact test for categorical data and the Mann-Whitney U test for continuous data. Logistic regression analyses assessed the association between the EuroLung1 score and major morbidity. To assess whether EuroLung1 provided prognostic information beyond established risk factors, a multivariable logistic regression model was constructed including EuroLung1 score, ppoDLCO%, and preoperative serum lactate dehydrogenase (LDH). These variables were selected based on previous studies demonstrating their independent association with postoperative morbidity following major lung resection (1,12). Model discrimination was evaluated by the area under the receiver operating characteristic curve (AUC); calibration was assessed using the Hosmer-Lemeshow goodness-of-fit test, Brier score and observed-to-expected ratio (O/E ratio). Missing data were handled using complete-case analysis. Statistical significance was set at P<0.05. Analyses were performed using IBM SPSS Statistics, version 22.0 (IBM Corp., Armonk, NY, USA).

Results

The cohort comprised 293 women and 351 men with a median age of 66 years (IQR 59–74 years). The surgical procedures included 574 lobectomies (89.1%) and 70 segmentectomies (10.9%). In 24 patients (3.7%), thoracoscopic procedures had to be converted to thoracotomy.

Histopathological evaluation revealed 548 patients with confirmed NSCLC (85.1%), 30 patients with pulmonary metastasis (4.7%) and 66 patients with benign disease (10.2%). Patient demographics and clinical characteristics are categorized in Table 1. The median EuroLung1 score in our cohort was 6 (IQR 3–9).

Table 1. Patient demographics and clinical characteristics.

Variable	Value (n=644)
Age (years)	66.0 (59.0–74.0)
Male	351 (54.5)
Smoker	436 (67.7)
Comorbid diseases
Coronary artery disease	102 (15.8)
Congestive heart failure	60 (9.3)
Chronic pulmonary disease	200 (31.1)
Peripheral vascular disease	63 (9.8)
Cerebrovascular disease	58 (9.0)
Diabetes	83 (12.9)
Moderate to severe renal disease	21 (3.3)
Moderate to severe liver disease	1 (0.2)
ECOG ≥2	13 (2.0)
ASA-PS >2	247 (38.4)
FEV1%	76.1 (66.7–85.0)
ppoFEV1%	61.5 (52.6–69.6)
DLCO%	67.3 (54.2–80.5)
ppoDLCO%	54.0 (42.4–64.8)
Pathologic stage (n=548)
Stage I	424 (77.4)
Stage II	82 (15.0)
Stage III	42 (7.7)

Data are presented as n (%) or median (IQR). ASA-PS, American Society of Anesthesiologists physical status; DLCO%, diffusing capacity of the lung expressed as the percent predicted; ECOG, Eastern Cooperative Oncology Group; FEV₁%, forced expiratory volume in the first second expressed as the percent predicted; IQR, interquartile range; ppo, predicted postoperative.

Six patients (0.9%) died in the postoperative period. Four patients died from postoperative pneumonia. One patient developed a bronchopleural fistula followed by sepsis and multiple organ failure and died on postoperative day 12. Another developed pneumonia and atrial arrhythmia on postoperative day 5 followed by a bronchopleural fistula, and died on postoperative day 79. Postoperative cardiopulmonary (i.e., major) complications occurred in 83 patients (12.9%). Overall morbidity including any complication or mortality was observed in 116 patients (18.0%). The distribution of individual complications is depicted in Table 2.

Table 2. Frequency of single postoperative complications.

Outcome	Value
Postoperative mortality	6 (0.9)
Pulmonary complications	59 (9.2)
Pneumonia	42 (6.5)
Atelectasis requiring bronchoscopy	16 (2.5)
Adult respiratory distress syndrome	7 (1.1)
Initial ventilator support for >48 hours	9 (1.4)
Unplanned re-intubation or tracheotomy	17 (2.6)
Cardiovascular complications	35 (5.4)
Acute myocardial infarction	2 (0.3)
Pulmonary embolism	1 (0.2)
Atrial arrhythmia	29 (4.5)
Ventricular arrhythmia	4 (0.6)
Other complications	73 (11.7)
Bronchopleural fistula	9 (1.4)
Empyema	18 (2.8)
Wound infection	2 (0.3)
Other postoperative infection	5 (0.8)
Cerebrovascular event	1 (0.2)
Delirium	5 (0.8)
Renal failure	4 (0.6)
Postoperative bleeding	13 (2.0)
Recurrent laryngeal nerve injury	12 (1.9)
Chylothorax	3 (0.5)
Other relevant events	11 (1.8)
Overall morbidity†	116 (18.0)

Data are presented as n (%). ^†, the occurrence of postoperative bleeding, chylothorax and recurrent laryngeal nerve injury was not included.

In univariable logistic regression analysis, the EuroLung1 score showed a statistically significant association with major morbidity: each 1-point increase in the score conferred 9% higher odds of complication [odds ratio (OR) =1.092, 95% confidence interval (CI): 1.010–1.180, P=0.03]. Similarly, advancing by one EuroLung1 risk class was associated with a 33% increase in odds (OR =1.327, 95% CI: 1.027–1.715, P=0.03). After adjustment for ppoDLCO% and preoperative serum LDH as additional independent predictors of morbidity the association between EuroLung1 score and major complications persisted, albeit with borderline significance (OR 1.079, 95% CI: 0.995–1.170, P=0.07). ROC analysis demonstrated poor discrimination (AUC =0.573). Calibration analysis demonstrated acceptable model fit according to the Hosmer-Lemeshow test (P=0.38). The Brier score was 0.1115, indicating acceptable overall predictive accuracy. The overall O/E ratio was 1.00, suggesting excellent calibration-in-the-large. However, calibration varied across EuroLung1 risk classes. The model tended to underestimate morbidity in lower-risk categories (classes I–II), whereas morbidity was generally overestimated in intermediate and higher-risk categories (classes III–V), indicating imperfect calibration across the risk spectrum (Table 3).

Table 3. Predicted and observed cardiopulmonary morbidity rates and O/E ratios across EuroLung1 risk classes.

Risk class	EuroLung1 score	Number of patients	Predicted morbidity (%)	Observed morbidity (%)	O/E ratio
I	0–1	41	5.2	7.3	0.94
II	2–4	163	8.2	10.5	1.06
III	5–7	221	14.3	12.2	0.98
IV	8–11	219	21.6	15.5	0.97
V	12–16	10	32.4	30.0	1.48

O/E, observed-to-expected.

Sensitivity analysis excluding the 24 patients who required intraoperative conversion to thoracotomy yielded similar results, with persistently poor discrimination of the EuroLung1 score (AUC =0.574) and no meaningful improvement in calibration (Hosmer-Lemeshow P=0.39). Additional sensitivity analyses restricted to patients with histologically confirmed NSCLC (n=548) and those undergoing lobectomy (n=574) yielded results comparable to those of the primary analysis. In the NSCLC subgroup, discrimination remained poor (AUC =0.573), calibration was acceptable (Hosmer-Lemeshow P=0.33, the Brier score was 0.1114, and the overall O/E ratio was 1.00. Similar findings were observed in the lobectomy subgroup (AUC =0.569, Hosmer-Lemeshow P=0.31, Brier score =0.1077, O/E ratio =1.00).

Notably, the observed complication rates in our cohort diverged from those predicted by the EuroLung1 model in the original publication (Figure 2). Specifically, the model systematically underestimated morbidity in lower-risk classes (EuroLung1 scores 0–7) and overestimated risk in higher-risk classes (scores ≥12).

Figure 2.

The predicted and observed cardiopulmonary morbidity rates stratified by the risk classes according to EuroLung1 score.

Discussion

Accurate preoperative risk assessment is essential to develop individualized treatment strategies and crucial to promote shared decision-making, particularly in high-risk patients. To date, the positive effect of minimally invasive approach on postoperative outcomes has not been mirrored in the risk stratification algorithms outlined in the current guidelines (3,4). In this context, the EuroLung scoring system aggregating multiple weighted predictors, including surgical approach (thoracotomy vs. minimally invasive approach) and key comorbidities offers a more nuanced, quantitative framework for preoperative risk stratification, particularly for patients undergoing minimally invasive lobectomy or segmentectomy.

In this external validation, the EuroLung1 score demonstrated a statistically significant association with major complications following minimally invasive anatomic lung resection. However, discrimination was limited. Regarding calibration, our analysis demonstrated that the EuroLung1 model tended to underestimate morbidity in lower-risk classes and generally overestimate morbidity in higher-risk classes. Moreover, calibration varied across risk classes. Additional sensitivity analyses restricted to patients without intraoperative conversion to thoracotomy, patients with NSCLC and those undergoing lobectomy yielded findings comparable to those observed in the full cohort, suggesting that the limited discrimination and imperfect calibration of the EuroLung1 model are not primarily attributable to the inclusion of patients with benign disease, pulmonary metastases, or segmentectomy. This result is in line with previously published data derived from a large institutional database in Austria consisting exclusively of minimally invasive anatomic lung resection and further underscores the weak correlation between predicted and observed outcomes (10).

The limited performance of the EuroLung1 score in our cohort may stem from several factors. First, overfitting is a plausible concern, as internal validation to assess overfitting was not reported in the original publication (5). Overfitting occurs when a regression model fits too closely to the training data, resulting in poor generalizability to independent datasets (17). Second, the exclusion of diffusing capacity (DLCO) and preoperative serum LDH as well-known independent predictors of morbidity after lung resection might have reduced the robustness of EuroLung model (12,18). Indeed, our multivariable analysis showed that EuroLung1 lost significance after adjusting for ppoDLCO% and LDH, underscoring the impact of omitted predictors. Furthermore, the ESTS derivation cohort included only 13.1% minimally invasive cases, whereas our cohort consisted exclusively of minimally invasive cases (5). Consequently, the risk-mitigating effect of the minimally invasive approach may have been underestimated in the original model. This highlights the need for regular model updates to better reflect contemporary surgical practice. Finally, and perhaps most importantly, our population reflects the typical German thoracic surgery case mix, enriched with elderly patients and those with impaired pulmonary function, for whom the minimally invasive approach may attenuate surgical risk beyond what EuroLung1 anticipates.

The limited performance of the EuroLung1 model in our cohort also highlights a general limitation of conventional risk scores based on a fixed set of predictors with predefined additive weighting. Although innovative approaches such as machine learning may offer additional flexibility in modelling complex and non-linear relationships, their potential advantages cannot be assessed from the present data (19,20). Therefore, a more immediate and pragmatic strategy may be recalibration or updating of existing thoracic surgical risk models, including incorporation of contemporary predictors and data derived from minimally invasive surgical cohorts.

The authors acknowledge the limitations of this study, which is subject to the inherent constraints of retrospective, single-center analyses. Furthermore, our cohort consisted exclusively of patients from a German tertiary center, limiting generalizability to populations with differing case mixes or practice patterns.

Conclusions

In conclusion, the EuroLung1 score remained significantly associated with major morbidity following minimally invasive anatomic lung resection; however, its discriminative performance was poor and calibration was imperfect across risk classes. These findings suggest limited performance of the model in a contemporary minimally invasive cohort and indicate that recalibration or updating may be required before its use for individual risk prediction.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was approved by the Ethics Committee of the Hannover Medical School (ID: 019/2016BO2), and the requirement for individual informed consent was waived due to the retrospective nature. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Data Sharing Statement

Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0821/dss

jtd-18-9-1047-dss.pdf ^{(67.6KB, pdf)}

DOI: 10.21037/jtd-2026-0821