The UK Biobank submaximal cycle ergometer test for assessment of cardiorespiratory fitness: Validity, reliability, and association with disease outcomes
Background: Cardiorespiratory fitness (CRF) was assessed in UK Biobank (UKB) using heart rate response to a submaximal ramped cycle ergometer test that was individualised for participant characteristics including cardiovascular disease risk. Studies have since explored health associations with CRF by estimating maximal oxygen consumption (VO2max) from UKB test data using interpretation methods that have not accounted for this individualisation procedure. Thus, dose-response relationships reported in these studies may be inaccurate. We developed and validated a novel VO2max estimation approach that accounts for the UKB test individualisation procedure and compared dose-response relationships with health outcomes between the novel and previous methods. Methods: In a cross-over study (n=189), participants completed several UKB tests and VO2max was measured. A multilevel modelling framework was developed that combines heart rate response features from the UKB test to estimate VO 2 max. Estimates were compared within participants across UKB test protocols, and with directly measured VO2max. Short-term test-retest reliability was assessed in a subsample of participants (n=87). In UKB, we examined associations between estimated CRF and disease endpoints (n=80,259) and compared associations obtained with an unvalidated method. Long-term test-retest reliability was examined (n=2877). Results: Estimated and directly measured VO2max were strongly correlated (Pearsons r range: 0.68 to 0.74) with no mean bias (women bias: -0.8 to 0.4; men bias range: -0.3 to 0.3), outperforming a previous approach for interpreting UKB test data. Agreement between estimated VO2max across different test protocols was strong (Pearsons r range: 0.94 to 0.99). Short- and long-term reliability was also high (lambda=0.91 and 0.80, respectively). All-cause mortality was 7 (95%CI 4-10%, 2686 deaths) lower and CVD mortality 9% (95%CI 3-14%, 858 deaths) lower for every 1-MET difference in fitness, associations twice as strong as determined by previous methods. Conclusions: We present a valid and reliable method for estimating CRF in UKB and demonstrate its utility in characterising dose-response relationships with health outcomes. Accounting for the individualisation procedure strengthens observed relationships between CRF and disease and enhances the case for promoting improved fitness in the general population.