We conducted a retrospective cohort study to evaluate the association between CRF, quantified by estimated metabolic equivalents (METs) on ETT, and all-cause mortality. The study population consisted of consecutive adult patients undergoing stress testing at our institution from January 1, 1991, through December 31, 2014. In cases of additional testing, only the first stress test was selected. Patients who underwent pharmacologic stress testing (n = 38 828) or who were converted to pharmacologic testing because of inability to reach 85% of maximum predicted heart rate (n = 467) were excluded. We excluded patients for whom sex information was missing (n = 796). Data analysis was performed from April 19 to July 17, 2018. The Cleveland Clinic Foundation Institutional Review Board approved the study and waived patient informed consent. Data were not deidentified.

At the time of stress testing, patient demographics (age and sex), anthropometrics (height, weight, and body mass index), medications, and comorbidities were prospectively documented. These comorbidities included history of CAD, diabetes, hypertension, hyperlipidemia, end-stage renal disease (ESRD), and smoking. Study definitions for comorbid conditions are provided in the eMethods in the Supplement. Data and analyses are presented in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies.²⁸

Patients underwent symptom-limited ETTs according to standardized protocols. The specific protocol for each test was chosen by the exercise physiologist supervising the test, based on the patient’s reported activity, and the test was performed as recommended by the exercise testing guidelines.²⁹ The CRF was quantified as peak estimated METs and was determined based on treadmill grade and speed at peak exercise. Within each sex, the distribution of achieved METs separated by decade of age was used to identify the 25th, 50th, 75th, and 97.7th percentiles of CRF. Patients were then stratified a priori into performance groups by age- and sex-associated cutoffs as follows: elite (≥97.7th percentile), high (75th-97.6th percentile), above average (50th-74th percentile), below average (25th-49th percentile), and low (<25th percentile).

Estimated METs were calculated for each patient using the Veterans Affairs cohort formula for men (estimated METs = 18.7 – [0.15 × age])³⁰ and the St James Take Heart Project formula for women (estimated METs = 14.7 – [0.13 × age]).⁵ In respective sexes, these formulas have been previously reported to perform best in terms of their ability to determine outcomes.³¹ Percentages of estimated METs were then calculated using the following ratio: achieved (METs/estimated METs) × 100.

The primary outcome was all-cause mortality and was determined from the Social Security Death Index, when available.³² It was supplemented by the institutional death index (medical record documentation of patient’s death), particularly for the period following November 2011, when restrictions for the Social Security Death Index access were implemented. The final censoring date was December 31, 2017.

Data are reported as mean (SD) for normally distributed variables and median (interquartile range) for nonnormally distributed, continuous variables. Analysis of variance and χ² testing were used for the analysis of continuous and categorical variables, respectively. The time-related all-cause mortality was analyzed using the nonparametric Kaplan-Meier method,³³ and groups were compared using the log-rank test.

To adjust for differences in baseline characteristics between performance groups, a multivariable Cox proportional hazards regression model was constructed to obtain the risk-adjusted association between all-cause mortality and CRF. Covariates incorporated into the model included age, sex, body mass index, history of CAD, hyperlipidemia, hypertension, diabetes, smoking, ESRD, year of testing, and current use of aspirin, β-blockers, or statins. We additionally found that patients referred for ETT for specified indications other than symptoms or those related to known or suspected CAD had significantly worse survival (eFigure 1 in the Supplement). These indications for testing were incorporated into the regression model.

For all analyses, a 2-sided P ≤ .05 was considered statistically significant. Analyses were performed April 2018 using JMP statistical software, version 13 (SAS Institute Inc).

A total of 122 007 patients (mean [SD] age, 53.4 [12.6] years; 72 173 [59.2%] male) were included in the final study cohort (eFigure 2 in the Supplement). Baseline characteristics are given in Table 1. The prevalence of associated comorbidities decreased significantly with increasing performance with the exception of hyperlipidemia, which was present in 31.6% (1128 of 3570) of elite performers and only 25.1% (7323 of 29 181) of low performers (P < .001). Temporal trends for age and comorbidities among the study cohort remained stable, although use of common cardiovascular medications increased during the study period (eFigures 3, 4, and 5 in the Supplement). Performance group cutoffs based on age and sex are given in Table 2. Evaluation for CAD and symptom assessment were the most common indications for ETT referral, with additional indications listed in the eTable in the Supplement.

Death from any cause occurred in 13 637 patients during a median follow-up of 8.4 years (range, 4.3-13.4 years) and 1.1 million person-years of observation. Kaplan-Meier curves of all-cause mortality stratified by performance groups demonstrated significant, incremental reduction in all-cause mortality associated with increasing performance (Figure 1). Elite performers had increased unadjusted survival compared with all other groups, including high performers.

To assess for potential confounding related to indications for ETT, survival analysis was performed by referral indications. Indications were grouped into 3 categories: symptoms only, known or suspected CAD, or other or non-CAD indications. Kaplan-Meier curves by indication groups demonstrated significantly worse survival in patients referred for other or non-CAD indications for ETT (eFigure 1 in the Supplement).

After multivariate adjustment, incremental reduction in all-cause mortality was associated with increasing performance when compared with low performers (Figure 2A). Similar findings were observed when separated by sex (Figure 2B). Adjusted hazard ratios (HRs) for clinical comorbidities and among performance groups are shown in Figure 2C. Risk-adjusted all-cause mortality was inversely proportional to CRF and was lowest in elite performers (elite vs low: adjusted HR, 0.20; 95% CI, 0.16-0.24; P < .001). The increase in all-cause mortality associated with reduced CRF (low vs elite: adjusted HR, 5.04; 95% CI, 4.10-6.20; P < .001; below average vs above average: adjusted HR, 1.41; 95% CI, 1.34-1.49; P < .001) was comparable to or greater than traditional clinical risk factors. Adjusted mortality risk was expectedly highest in the lowest performing groups and generally exceeded that of traditional clinical risk factors, including CAD (adjusted HR, 1.29; 95% CI, 1.24-1.35; P < .001), smoking (adjusted HR, 1.41; 95% CI, 1.36-1.46; P < .001), and diabetes (adjusted HR, 1.40; 95% CI, 1.34-1.46; P < .001). Multivariate Cox proportional hazards regression with less stringent (≥95th percentile) and more stringent (≥99th percentile) elite performance cutoffs demonstrated consistent findings (eFigure 6 in the Supplement).

Multivariate analyses also demonstrated improved survival in elite vs high performers (adjusted HR, 0.77; 95% CI, 0.63-0.95; P = .02). By sex, a nonstatistically significant improved survival in elite vs high performers was present in both men and women (men: adjusted HR, 0.81; 95% CI, 0.64-1.03; P = .09; women: adjusted HR, 0.65; 95% CI, 0.41-1.02; P = .06) (Figure 3). When analyzed by age group, the difference in survival between elite and high performers was only maintained in older patients. There was a statistically significant reduction in mortality between elite and high performers 70 years or older (adjusted HR, 0.71; 95% CI, 0.52-0.98; P = .04). In younger age groups, there was no difference in survival between elite and high performers.

Further multivariate regression was performed in the following clinical subgroups: known CAD, diabetes, hypertension, and hyperlipidemia. In all comorbid subgroups, all-cause mortality was inversely proportional to CRF and lowest in elite performers (eFigure 7 in the Supplement). In patients with hypertension, elite performers had a significant reduction in risk-adjusted all-cause mortality compared with high performers (adjusted HR, 0.70; 95% CI, 0.50-0.99; P = .05). In all other comorbid subgroups, there was no statistical difference in survival between elite and high performers.

The primary limitation of the study reflects its retrospective nature, in that the association between CRF and mortality does not prove causation. The degree to which high CRF preselects patients with lower mortality vs causes a reduction in mortality is not discernible from our study. In addition, although we account for many clinical factors, there may be confounding variables that are unaccounted for in our analyses, including socioeconomic status, race/ethnicity, and others. These limitations apply to other studies of aerobic fitness and long-term outcomes and are unlikely to be overcome because of the infeasibility of a large-scale, randomized study. Given these inherent limitations, our study demonstrates that patients who can exercise to an extreme level live longest, acknowledging that there may be many measured and unmeasured factors that contribute to this association.

We also recognize that the study population (ie, those referred for ETT) may not reflect the general population distribution of estimated functional capacity for the purpose of identifying elite performers. However, there is no widely accepted standard for classification of CRF based on ETT performance. Recent proposals have suggested similar age-specific methods for CRF standardization.⁴⁰ The method outlined in the present study adequately defines a cohort of patients with exceptionally high CRF suitable to test the underlying hypothesis. With the use of previously validated models to estimate functional capacity in adults,^5,30 the elite performance group in this study had a mean estimated functional capacity 55% greater than estimated by age and sex. Cutoff values for elite performers were consistent with historic normative data,⁴¹ and measured CRF of elite performers was comparable to published age- and sex-matched cohorts of endurance athletes^42,43,44

Cardiorespiratory fitness was based on a patient’s performance on a single ETT, and these findings do not speak to the association between long-term levels of fitness and mortality. However, this study reinforces the clinical utility of using ETT to better understand and determine patient prognosis. Even measured at a single time point, performance on ETT is remarkably correlated with long-term survival.