Comparative performance of mortality prediction using different reference value equations for peak oxygen uptake in cardiopulmonary exercise testing

Background Cardiorespiratory fitness (CRF) is a well-established health indicator with significant prognostic value for all-cause mortality. Peak oxygen uptake (VO2peak) measured through cardiopulmonary exercise testing (CPET) is considered the gold standard method. As VO2peak is influenced by sex and age, it is compared to reference standards, and the percent-predicted value achieved (%VO2peak) is considered a more reliable indicator for individual assessment. However, various reference value equations for VO2peak exist, demonstrating international heterogeneity. Additionally, the new Brazilian prediction equation (BE) has not been evaluated for prognostic performance, and the cut-off value for risk stratification remains unknown. Purpose To compare the predictive performance for mortality stratification of three VO2peak reference value equations: Wasserman[1], FRIEND[2], and the Brazilian equation[3]. Methods A cohort of patients underwent CPET at a private center in Rio de Janeiro state, Brazil, between May 2018 and January 2023. Survival status was determined using death certificates. CRF was assessed by %VO2peak for each prediction equation. Prognostic performances were evaluated using the area under the curve (AUC) of the receiver operating characteristic curve (ROC). The standard error of AUC was calculated using the method of Hanley & McNeil. The optimal cut-off point for %VO2peak in each reference value equation was determined based on the ROC curve's ideal combination of sensitivity and specificity for predicting mortality. Results A total of 2,684 individuals (62.4% men; 52.9 ± 14.5 years) were included, with a median follow-up of 451 days (IQR = 422). During this period, 31 deaths (1.15%) were identified. The AUC was 0.812 (95%CI: 0.743 - 0.882) for Wasserman, 0.796 (95% CI: 0.710 - 0.882) for FRIEND, and 0.753 (95% CI: 0.661 - 0.846) for Brazilian equation. The AUC for Wasserman was significantly higher than that of Brazilian equation (p = 0.018). No differences were observed between Wasserman and FRIEND, and between FRIEND and Brazilian equation. The optimal cut-off points were: %VO2peak ≤ 84% [Sens: 80.7%; Spec: 69.4%) for Wasserman; %VO2peak ≤ 82% [Sens: 77.4%; Spec: 57.3%] for FRIEND; and %VO2peak ≤ 76% [Sens = 80.7%; Spec = 57.75%] for Brazilian equation. Conclusions In this sample, all reference value equations demonstrated good performance in predicting mortality. The Brazilian equation, although developed for treadmill assessment, exhibited similar accuracy to the FRIEND equation and slightly lower accuracy than Wasserman's equation. The lower cut-off value for the Brazilian equation may be attributed to ergometer specificity, as reference value are generally higher for treadmills.