Background: Cardiopulmonary exercise testing (CPET) enables measurement of the slope of the increase in minute ventilation in relation to carbon dioxide elimination during exercise (the VE/VCO2 slope). Several studies have shown that the VE/VCO2 slope is a strong marker for postoperative complications and mortality. However, current thresholds for adverse outcomes are generated from historical data in heart failure patients. Methods: This was a retrospective analysis of 158 patients with lung cancer who underwent lobectomy or pneumonectomy during 2008-2020. The main outcome was major pulmonary complications (MPC) or death <= 30 days of cancer surgery. Patients were first categorized using two different single threshold approaches; the traditional threshold of 35 and the highest Youden value from the receiver operating curve (ROC) analysis. Secondly, patients were categorized into three risk groups using two thresholds. These two thresholds were determined in an ROC analysis, where the VE/VCO2 slope values generating either a 90% sensitivity (lower threshold) or a 90% specificity (upper threshold) for the main outcome were chosen. The frequency of complications was compared using Chi(2). The overall model quality was evaluated by an area under the curve (AUC) analysis. Positive predictive values (PPVs) and negative predictive values (NPVs) are presented. Results: The two thresholds, <30 (90% sensitivity) and >41 (90% specificity), created three risk groups: low risk (VE/VCO2 slope <30, n=44, 28%); intermediate risk (VE/VCO2 slope 30-41, n=95, 60%) and high risk (VE/VCO2 slope >41, n=19, 12%). The frequency of complications differed between groups: 5%, 16% and 47% (P<0.001). Using two thresholds compared to one threshold increased the overall model quality (reaching AUC 0.70, 95% confidence interval: 0.59-0.81), and identified a high sensitivity threshold (VE/VCO2 slope <30) which generated a NPV of 95% but importantly, also a high specificity threshold (VE/VCO2 slope >41) with a PPV of 47%. Conclusions: Risk stratification based on three risk groups from the preoperative VE/VCO2 slope increased the model quality, was more discriminative and generated better PPV and NPV compared to traditional risk stratification into two risk groups.
