Importance of Treadmill Exercise Time as an Initial Prognostic Screening Tool in Patients with Systolic Left Ventricular Dysfunction

Abstract

Background

We sought to determine if treadmill exercise time may be of value as an initial prognostic screening tool in ambulatory patients with impaired systolic function referred for cardiopulmonary exercise testing.

Methods and Results

We studied 2,231 adult systolic heart failure patients (27% women) who underwent cardiopulmonary stress testing using a modified Naughton protocol. We assessed the value of treadmill exercise time for prediction of all-cause death and a composite of death or UNOS status 1 heart transplantation. During a mean follow up of 5 years, 742 (33%) patients died. There were 249 (11%) UNOS status 1 heart transplants. Treadmill exercise time was predictive of death and the composite outcome in both women and men, even after accounting for peak oxygen consumption and other clinical covariates (adjusted hazard ratio of lowest versus high sex-specific quartile for prediction of death 1.70, 95% CI 1.05–2.75, P=0.03, and for prediction of the composite outcome 1.75, 95% CI 1.15–2.66, P=0.009). For a one minute change in exercise time there was a 7% increased hazard of death (e.g. comparing 480 to 540 seconds HR 1.07, 95% CI 1.02–1.12, P=0.004).

Conclusions

Since cardiopulmonary stress testing is not available in every hospital, treadmill exercise time using a modified Naughton protocol may be of value as an initial prognostic screening tool.

Peak oxygen consumption (VO₂) remains one of the most powerful single predictors of mortality for heart failure (HF) patients with severe systolic left ventricular dysfunction^1–3. Current guidelines suggest ambulatory patients be considered for transplantation when the peak VO₂ ≤ 14 ml/kg/min or ≤ 12 ml/kg/min in the setting of β-blockade³. Although cardiopulmonary exercise testing is routinely used to determine candidacy for heart transplantation, it is not known whether a simple measurement of treadmill exercise time may be of comparable value as an initial prognostic screening tool. We sought to evaluate whether (1) treadmill exercise time predicts survival in HF patients with systolic dysfunction, (2) treadmill exercise time adds incremental prognostic value beyond established risk factors including peak VO₂ and (3) a model with treadmill exercise time in addition to established risk factors better classifies risk than a model with established risk factors alone.

Methods

We performed an observational prospective cohort study at the Cleveland Clinic involving consecutive patients with left ventricular ejection fraction < 40% who underwent cardiopulmonary stress testing between August 1997 to April 2007. We focused specifically on patients who were tested according to the modified Naughton protocol, which was the most common protocol used in our laboratory for this type of patient. Patients were excluded if they were < 18 years old or had no United States Social Security number. The study was approved by the institutional review board at the Cleveland Clinic, and since all data were collected and recorded as part of routine clinical care the requirement for informed consent was waived.

Clinical Data

All demographic information, medications, medical and surgical history, heart rate, blood pressure, directly measured height and weight, and stress results were obtained at the time of stress testing and prospectively recorded in our electronic database. Biventricular pacemakers were not reported separately from other pacemakers during database entry. Glomerular Filtration Rate (GFR) was estimated using the Cockcroft-Gault equation⁴. Left ventricular ejection fraction was determined either by echocardiography, left ventriculography, or ECG-gated myocardial perfusion imaging. If a patient underwent more than one metabolic stress test using a modified Naughton protocol, only the first test was considered. Serum laboratory tests within 3 months of the cardiopulmonary study were included and only the laboratory tests closest to the stress test date were used.

Cardiopulmonary stress testing

Cardiopulmonary stress testing was symptom limited and patients were strongly advised to not use the handrails for support. Details regarding our laboratory’s protocols have been previously published⁵. Briefly, the test was performed using a modified Naughton protocol which increases workload about 1 metabolic equivalent every 2 minutes⁶. Results were recorded on a MedGraphics cardiopulmonary system (St. Paul, Minnessota). Oxygen consumption, carbon dioxide production, minute ventilation, tidal volume, heart rate, blood pressure, and respiratory rate were obtained at rest, every 30 seconds during exercise, and during recovery. Total duration of exercise, heretofore referred to as treadmill exercise time, was recorded to the nearest second.

End Points

The primary endpoints were all-cause mortality and a composite of all-cause mortality or United Network for Organ Sharing (UNOS) Status 1 heart transplantation during a mean follow-up of 5 years (maximum for survivors, 11 years). Mortality data was obtained by linking our database with the United States Social Security Administration Death Index. We previously validated this approach, yielding a sensitivity of 97% ⁷, which is equivalent to a 97% follow-up rate. UNOS status 1 heart transplantation was based on Organ Procurement and Transplantation Network data as of June 18, 2008. All events (death or heart transplantation) were censored as of April 1, 2008.

Statistical Analysis

Baseline characteristics were reported as sex-specific quartiles of treadmill exercise time. Continuous variables were expressed as means with standard deviations and categorical variables were expressed as frequencies. Serum laboratory tests prior to October 1999 were systematically unavailable on our electronic database. We used informed imputation to fill in these missing values by constructing regression models for each laboratory measure on all other characteristics of our existing database except for outcome (10% of creatinine clearance, BUN, glucose, and sodium were imputed and 15% of serum hemoglobin values were imputed).

We generated Kaplan-Meier plots relating treadmill exercise time and cumulative mortality. We constructed unadjusted and adjusted Cox proportional hazards model to examine the association of treadmill exercise time and outcomes. Possible nonlinear associations were tested with restricted cubic splines. Potentially important interactions were tested. The proportional hazards assumption was tested by scaled Schoenfeld residuals and inspection of hazard ratio plots.

Model discrimination was assessed by calculating the out-of-bag c-index for time to event outcomes, an approach that we have used previously⁸. The out-of-bag (OOB) method involves obtaining bootstrap samples from the original cohort and using each sample to compute a prediction model. Each bootstrap sample left out about 37% of the data, which was referred to as the OOB data. The prediction model was applied to the OOB data to calculate the OOB c-index, a measure that is conceptually similar to the area under a receiver operating curve (AUC under ROC). The purpose of this method is to perform multiple internal validations with an internal cohort that has been randomly sampled multiple times. This is a more conservative measure than what we have reported in previous work which was based on one assessment of the original data⁹. Hence, we would expect that these OOB c-index values will be lower than what we previously reported, as they represent more conservative estimates.

To determine the change in prediction error attributable to each variable, we recalculated the prediction error after random permutation of that variable in the OOB data; a variable with a high degree of importance would be expected to yield a greater change in the OOB c-index. The process was repeated 100 times for each variable.

To address clinical utility we constructed risk reclassification figures. We compared 1-year predicted risk estimates based on models of established risk factors with and without variables in question. We picked categories of risk of 15% based on the current 1-year mortality outcomes after cardiac transplantations¹⁰. We used two objective ways of quantifying improvement in categories as suggested by Pencina: net reclassification improvement (NRI) and integrated discrimination improvement (IDI)¹¹. The NRI and IDI were calculated by comparing individual patient’s predicted risks at 1-year based on a Cox model 1 vs. another Cox model 2. These predicted risks were calculated using a modification of the R function  predict() with censoring at 1-year.

Statistical analysis was performed using the SAS version 9.1.3 (SAS Institute Inc., Cary, NC) and R version 2.6.2 (www.R-project.org). We used Harrell’s Design and Hmisc libraries for model construction and assessment, an R macro written by one of the authors (H.I.) for the OOB and change in prediction error analyses, and another R macro written by Katan and colleagues for calculating the observed risks in the reclassification figures. We used a SAS macro written by Pencina for calculating the NRI and IDI ¹¹.

This work was supported in part by Health Resources and Services Administration contract 234-2005-370011C. The content is the responsibility of the authors alone and does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. government. The authors had full access to the data and take responsibility for the integrity of the data. All authors have read the manuscript and approved it for submission.

Results

Our cohort consisted of 602 (27%) women and 1629 (73%) men. During a mean follow up of 5 years (maximum for survivors, 11 years), 155 women (26% of female cohort) and 587 (36% of male cohort) men died. There were 249 patients (16% women) who underwent heart transplantation as UNOS status 1. Only four patients were transplanted twice.

Table 1 shows the baseline characteristics of the 2231 patients according to sex-specific quartiles for treadmill exercise time. Compared to those in the highest quartile, those in the lowest quartile were older, more likely to have coronary artery disease, diabetes, implantable cardioverter-defibrillators, pacemakers, and had lower systolic blood pressures with higher resting heart rates. They also had lower peak VO₂ and lower peak respiratory exchange ratio. They were less likely to be on a β-blocker, or an angiotensin converting enzyme inhibitor and more likely to be on nitrates, hydralazine, angiotensin receptor blockers, diuretics and antiarrythmic agents. Laboratory tests were similar except those in the lowest quartile had higher BUN and lower creatinine clearance.

Table 1.

Baseline Characteristics According to Sex-Specific Quartiles for Treadmill Exercise Time

	Quartiles 1	Quartile 2	Quartile 3	Quartile 4	P
	(N=553)	(N=588)	(N = 524)	(N = 566)
Treadmill Time: Women (N=602) (range,sec)	21–315	317–478	480–600	603–1320
Treadmill Time: Men (N=1629) (range,sec)	35–357	360–480	484–657	660–1415
Age, (yrs)	57 +/− 10	56 +/− 10	54 +/− 11	49 +/− 11	<0.001
Body mass index, (kg/m2)	29 +/− 6	28 +/− 6	29 +/− 6	28 +/− 5	0.20
Current smokers, n (%)	112 (20)	117 (20)	108 (21)	122 (22)	0.91
Diabetes: insulin treated, n (%)	74 (13)	78 (13)	42 (8)	21 (4)	<0.001
Diabetes: not insulin treated, n (%)	111 (20)	107 (18)	84 (16)	48 (8)	<0.001
Coronary artery disease, n (%)	267 (48)	278 (47)	210 (40)	151 (27)	<0.001
Previous MI, n (%)	68 (12)	63 (11)	76 (15)	72 (13)	0.30
Previous CABG, n (%)	187 (34)	192 (33)	131 (25)	84 (15)	<0.001
Previous PCI, n (%)	135 (24)	147 (25)	123 (23)	71 (13)	<0.001
Implantable cardioverter-defibrillator, n (%)	209 (38)	180 (31)	159 (30)	99 (17)	<0.001
Pacemaker, n (%)	160 (29)	149 (25)	117 (22)	76 (13)	<0.001
Medication use, n (%)
β-Blocker	349 (63)	357 (61)	345 (66)	378 (67)	0.13
Ace inhibitor	378 (68)	443 (75)	422 (81)	468 (83)	<0.001
Angiotensin receptor blocker	86 (16)	92 (16)	56 (11)	56 (10)	0.003
Potassium sparing diuretics	181 (33)	186 (32)	152 (29)	130 (23)	0.001
Antiarrhythmic	159 (29)	152 (26)	117 (22)	81 (14)	<0.001
Anticoagulation	253 (46)	255 (43)	219 (42)	172 (30)	<0.001
Aspirin	261 (47)	270 (46)	251 (48)	256 (45)	0.81
Digoxin	376 (68)	425 (72)	361 (69)	408 (72)	0.28
Nitrates	212 (38)	223 (38)	162 (31)	142 (25)	<0.001
Hydralazine	44 (8)	40 (7)	29 (6)	23 (4)	0.041
Loop diuretics	502 (91)	530 (90)	447 (85)	401 (71)	<0.001
Thiazide diuretics	90 (16)	98 (17)	53 (10)	38 (7)	<0.001
Statin	210 (38)	266 (45)	187 (36)	187 (33)	<0.001
Calcium channel blocker:
Non-dihydropyridine	6 (1)	2 (0)	4 (1)	4 (1)	0.52
Calcium channel blocker: dihydropyridine	24 (4)	33 (6)	31 (6)	11 (2)	0.005
Resting heart rate, (beats/min)	80 +/− 14	78 +/− 14	75 +/− 14	72 +/− 14	<0.001
Resting systolic blood pressure, (mm Hg)	108 +/− 18	110 +/− 18	111 +/− 18	112 +/− 17	<0.001
Left ventricular ejection fraction, (%)	19 +/− 8	20 +/− 7	21 +/− 8	21 +/− 7	<0.001
Peak oxygen consumption, (ml/kg/min)	11 +/− 3	14 +/− 2	17 +/− 3	22 +/− 4	<0.001
Peak respiratory exchange ratio	1.03 +/− 0.16	1.07 +/− 0.11	1.10 +/− 0.09	1.11 +/− 0.07	<0.001
Serum Sodium (mmol/L)	139 +/− 4	139 +/− 3	140 +/− 3	140 +/− 3	<0.001
Creatinine Clearance (ml/min)	78 +/− 37	82 +/− 41	94 +/−43	110 +/− 44	<0.001
Serum BUN (mg/dL)	31 +/− 16	28 +/− 13	23 +/− 10	19 +/− 7	<0.001
Serum hemoglobin (g/dL)	13 +/− 2	13 +/− 1	14 +/− 1	14 +/− 1	<0.001
Serum glucose (mg/dL)	112 +/− 44	115 +/− 46	111 +/− 47	100 +/− 30	<0.001

N= total number of women and men per quartile

Outcomes

Figure 1 shows the association of sex-specific quartiles of treadmill exercise time with all-cause mortality. The composite of death and UNOS status 1 heart transplant yielded similar sex-specific results (Appendix Figure 1). In both women and men treadmill exercise time predicted survival (death or death/UNOS status 1 transplantation) with the worst survival in the quartile with the lowest treadmill exercise time. After adjusting for all the variables listed in Table 1 including peak VO₂ in a Cox proportional hazards model, a shorter exercise time remained associated with an increase hazard of death (Table 2) as well as death and UNOS status 1 heart transplantation (Appendix Table 1). For a one minute change in exercise time there was a 7% increased hazard of death (e.g. for a comparison of 480 to 540 seconds HR 1.07, 95% CI 1.02–1.12, P=0.004). To determine if the imputed laboratory values affected our data, we performed a sensitivity analysis leaving out all laboratory variables and found similar results (HR 1.08, 95% CI 1.03–1.13, P 0.002).

Figure 1.

Kaplan-meier plots for (A) women and (B) men, stratified by quartiles of treadmill exercise time. (A) Women (N=602) Quartile 1: 21–315 seconds, Quartile 2: 317–478 seconds, Quartile 3: 480–600 seconds, Quartile 4: 603–1320 seconds. (B) Men (N=1629) Quartile 1: 35–357 seconds, Quartile 2: 360–480 seconds, Quartile 3: 484–657 seconds, Quartile 4: 660–1415 seconds.

Table 2.

Treadmill Exercise Time and Outcome: Cox Proportional Hazards Analyses

Model: Death	Hazard Ratio (95% CI)	P
Treadmill exercise time as a continuous variable*
Unadjusted	2.23 (2.00 to 2.49)	<.0001
Adjusted for age and sex	2.23 (1.98 to 2.50)	<.0001
Adjusted for age, sex, history of CAD, and peak VO2	1.54 (1.23 to 1.92)	.0005
Multivariable adjusted	1.40 (1.12 to 1.76)	.004
Treadmill exercise time as a dichotomous variableˆ
Unadjusted	4.91 (3.86 to 6.24)	<.0001
Multivariable adjusted	1.70 (1.05 to 2.75)	.03

^*Comparisons are between the 25^th percentile (317 seconds in women, 360 seconds in men) and 75^th percentile (600 seconds in women, 657 seconds in men)

^ˆComparisons between quartile 1 (<316 seconds in women, <358 seconds in men) and quartile 4 (>602 seconds in women, >659 seconds in men)

We further evaluated the association of treadmill exercise time and survival for the subgroup of patients with peak VO2 ≥ 14 ml/kg/min. Even among these lower-risk patients, lower treadmill exercise time predicted worse outcomes in both women and men (Figure 2 and Appendix Figure 2). In this subset of patients with high peak VO₂, a one minute change in exercise time also yielded a 7% increased hazard of death in a fully adjusted model (i.e., for a comparison of 480 to 540 seconds HR 1.07, 95% CI 1.00–1.14, P=0.04).

Figure 2.

Kaplan-meier plots for (A) women and (B) men with peak VO₂≥14, stratified by tertiles of increasing treadmill exercise time. (A) Women: Tertile 1: < 518 seconds, Tertile 2: 518–639 seconds, Tertile 3: ≥ 640 seconds. (B) Men: Tertile 1: < 511 seconds, Tertile 2: 511–682 seconds, Tertile 3: ≥ 683 seconds. Peak VO₂₌ peak oxygen consumption

The c-index of the models containing all variables including exercise time were 0.723 for all-cause death and 0.744 for composite of death and UNOS status 1 heart transplantation. The out-of-bag (OOB) c-indexes, a more conservative measure, of the models containing all variables including treadmill exercise time were 0.698 (95% CI 0.670 to 0.726) for all-cause death (Table 3) and 0.727 (95% CI 0.703 to 0.751) for the composite outcome of death and UNOS status 1 heart transplantation (Appendix Table 2), indicating moderate discriminatory ability. There were no notable differences between the models that contained clinical variables and either peak VO₂ or treadmill exercise time or both. Treadmill exercise time was amongst the top three most important contributors to discriminative prediction in both sets of models (Figure 3 and Appendix Figure 3).

Table 3.

Out of Bag (OOB) Concordance Index Values For Cox Regression Models

	All-cause death
		OOB Concordance Index (95% Confidence Interval)
Clinical variables only	0.672	(0.646 to 0.698)
Peak VO2 only	0.675	(0.662 t0 0.688)
Time only	0.666	(0.651 to 0.681)
Peak VO2 + Time only	0.679	(0.666 to 0.692)
Clinical variables + Peak VO2	0.699	(0.673 to 0.725)
Clinical variables + Time	0.697	(0.671 to 0.723)
Clinical variables + Peak VO2 + Time	0.698	(0.670 to 0.726)

Figure 3.

Change in out-of-bagging determined prediction error. Only the 10 most important variables are shown. Results are based on 100 bootstrapped samples.

To better compare model performance within clinical categories, we classified patients into low-risk (<15%) and high-risk (≥15%) categories of 1-year risk, with these cut-points based on the observed national 1-year survival after heart transplantation. We compared models with and without treadmill exercise time by cross-classifying predicted risks, stratified by whether or not the patients died (Figure 4), or developed the composite end-point of death or transplantation (Appendix Figure 4). If an additional variable were to add no predictive value, all points would fall on the line of identity. Spread around the line indicates modulation of predicted risk; if the variable correctly modulates predicted risk, there should be a greater preponderance of events (red open circles) above the line of identify. Clinical variables and usage of peak VO₂ (Figure 4A) or treadmill exercise time (Figure 4B) improved net reclassification (NRI) for the outcome of death and the composite outcome (Appendix Figure 4A–B). However, a particularly noteworthy finding is that there were no significant differences in net reclassification between a model with clinical variables and treadmill exercise time versus clinical variables and peak VO_2. (Figure 4C and Appendix Figure 4C).

Figure 4.

Reclassification: One year low-risk (<15%) and high-risk (≥15%) survival categories are cut-offs based on the observed national 1-year survival after heart transplantation. The endpoint is death and improvement of classification is expressed as net reclassification improvement (NRI) which is dependent on the pre-defined cut-offs and integrated discrimination improvement (IDI) which is not limited to pre-determined risk categories. If a variable adds no predictive value to the model, all points fall on the dark line of identity within each figure. Spread around the line indicates modulation of predicted risk; if the variable correctly modulates predicted risk, there should be a greater preponderance of events (red open circles) above the line of identify. A. Model with clinical variables and peak VO₂ compared to model with only clinical variables. Peak VO₂ with cardiac risk factors improved classification in 74 patients (23 who died, 51 who survived) but worsened it in 71 patients (4 who died, 67 who survived) with an IDI of 2.4% B. Model with clinical variables and treadmill exercise time compared to model with only clinical variables. Treadmill exercise time with cardiac risk factors improved classification in 70 patients (21 who died, 49 who survived) but worsened it in 68 patients (4 who died, 64 who survived) with an IDI of 1.9% C. Model with clinical variables and treadmill exercise time compared to model with clinical variables and Peak VO₂ improved classification in 43 patients (6 who died, 37 who survived) but worsened it in 46 patients (8 who died, 38 who survived) with an IDI of −0.4% which was not statistically significant

The integrated discrimination improvement (IDI) (a reclassification measure that is not limited to pre-determined categories of risk) for a model with treadmill exercise time was 1.9% (P<0.0001) (Figure 4B) for the outcome of death and 3.4% (P<0.001) (Appendix Figure 4B) for the composite outcome. There were no significant difference in IDI between a model with clinical variables and treadmill exercise time versus clinical variables and peak VO_2. (Figure 4C and Appendix Figure 4C).

Discussion

In a large cohort of patients with impaired left ventricular systolic function who underwent the same cardiopulmonary stress testing protocol, we found that treadmill exercise time predicts survival and yields similar prognostic value to peak oxygen consumption (peak VO₂). Even among low risk patients with a peak VO₂ > 14 ml/kg/min those who had a lower treadmill exercise time had markedly worse outcomes. Our findings are consistent with the hypothesis that exercise capacity is a valuable initial prognostic screening tool in patients with systolic left ventricular dysfunction.

Our study extends upon previous work demonstrating the prognostic power of exercise capacity^{12, 13}. Exercise capacity (i.e. treadmill time or “estimated” METs) is predictive of survival in both healthy adults and those with known coronary artery disease^14–16. Exercise tolerance reflects a number of prognostically important factors including cardiac function, endothelial function, pulmonary function, oxygen carrying capacity, and autonomic nervous system balance. However, its usefulness as a prognostic screening tool in patients with systolic left ventricular dysfunction is less established.

Treadmill exercise time has not been used in most heart failure survival models^{1, 17–19} but it has been previously explored in at least three heart failure studies. In the Captopril-Digoxin Research Group report, exercise time on a modified Naughton protocol was deemed to have no significant prognostic value. However, the cohort was relatively healthy as patients who exercised less than 240 seconds were excluded; the average exercise time was quite high at 558 seconds²⁰. In a small study of 60 heart failure patients there was a high degree of correlation between treadmill exercise time and peak VO₂. However, due to the small sample size, this study did not have the power to determine prognostic significance²¹. More recently, a cohort of 31 children with idiopathic cardiomyopathy who underwent cardiopulmonary stress testing using a modified Naughton protocol were followed prospectively for a median follow-up of 1282 days and 20 of them died or underwent heart transplantation. Both univariable and multivariable analyses showed treadmill exercise time to be predictive of outcome²².

Can a simple exercise stress test replace cardiopulmonary stress testing for patients being evaluated for heart transplantation? Possibly but we do not advocate this approach. By both OOB change in c-index and IDI, we found that peak VO₂ is a more powerful predictor of mortality than treadmill exercise time. Other researchers have also found measured peak VO₂ to be a better predictor than either estimated peak VO₂ or work rate^{23, 24}. Another advantage of cardiopulmonary stress testing includes the ability to measure patient effort (i.e. respiratory exchange ratio). When patients achieve an adequate level of effort, peak VO₂ is a reliable and reproducible measurement and does not depend on utilizing only one exercise protocol. Exercise time may not be as reproducible based on the results of Heart Failure and A Controlled Trial Investigation Outcomes of exercise training (HF-ACTION) which repeated within 7 days a baseline exercise study in 401 systolic heart failure patients (87% used a modified Naughton protocol) and noted that the peak VO₂ was unchanged but exercise time increased by 26 seconds on average²⁵. Cardiopulmonary stress testing can also help identify non-cardiac causes for shortness of breath. However, cardiopulmonary stress testing is not available in every hospital and therefore treadmill exercise time may be a reasonable screening tool for physicians to determine prognosis and need for referral to advanced heart failure treatment centers. Based on our data, less than 5 minutes and 17 seconds for females and less than 6 minutes for males may be useful treadmill exercise time cut-off values when using a modified Naughton protocol since in our cohort this corresponded with about a 15% one year mortality. Future research in other cohorts will be needed to verify our findings.

The strengths of our study include a large sample size with a large number of hard events, prospective data collection, adequate power to present sex-specific results^{26, 27}, and the recording of treadmill exercise time to the nearest second. All endpoints were determined by query of databases outside of our center. Deaths were determined by Social Security files, whereas UNOS 1 transplantation was determined by direct query of UNOS files. Of note, 20% of patients who experienced UNOS 1 transplantation were transplanted outside of our center. Thus, an ascertainment bias based on only considering events known to Cleveland Clinic information systems was avoided. Body mass index was calculated based on direct measures of height and weight. We adjusted for over 35 clinical/demographic variables which included sex, peak VO₂, and coronary artery disease and still found a longer exercise time interval to be associated with a lower hazard of death. We also used both traditional (C index) and contemporary statistical discrimination and reclassification methods¹¹ to assess model performance.

The C-index values reported in this manuscript are lower than what we reported earlier regarding a similar cohort⁹. We now have adopted a more sophisticated and conservative method (OOB c-index) than what we used previously. This approach effectively simulates numerous internal validations and as expected model performance would be worse if applied to the entire original dataset.

Limitations include that this is a single center study with only patients who could ambulate on a treadmill. Since the cohort includes patients from 1997, there are few patients using aldosterone antagonists, modest β-blockers usage, and about 25% with an ICD. However, these all compare favorably to what is seen in registries recently published^28–31. Biventricular pacemakers were not reported separately during database entry but most are identified in the ICD category since at our institution biventricular pacemakers are almost always implanted with an ICD. Therefore, when adjusting for all possible confounding variables we are including those with an ICD alone and with an ICD combined with a biventricular pacemaker. Laboratory tests like B-type natriuretic peptides were not included in our baseline characteristics since not routinely obtained at our center between 1997 and 2007. Although BNP has prognostic value it also has not been used in other heart failure survival models^{17, 32–35}. Like all epidemiologic studies we also cannot account for variables that change with time that may impact on death such as patients receiving an ICD after the cardiopulmonary stress test or changes in medication with time. However, this limitation is inherent in all studies that have confirmed the value of cardiopulmonary stress testing as a prognostic tool. We also did not compare treadmill exercise time to indices of ventilatory inefficiency such as the VE/VCO₂ slope, PetCO₂, oscillatory ventilation, and oxygen uptake efficiency slope which are more powerful risk predictors when compared with peak VO₂. These indices require minute-by-minute measurements which was not available in our database.

Despite these limitations, we found that treadmill exercise time is a powerful and independent predictor of outcomes in both women and men with impaired left ventricular systolic function, even after accounting for peak oxygen consumption. These findings have potentially important implications for our understanding of the clinical pathophysiology of systolic cardiac dysfunction as well as for finding optimal pathways to screen heart failure patients for possible transplantation. Since cardiopulmonary stress testing is not available in every hospital, treadmill exercise time using a modified Naughton protocol may be of value as an initial prognostic screening tool.

Physicians often refer patients with advanced heart failure for exercise testing with metabolic gas exchange measurements in order to assess suitability for cardiac transplantation. Peak oxygen consumption and other metabolic measures are known to be powerful predictors of mortality. We asked whether treadmill exercise time according to a standardized protocol might be comparable to peak oxygen consumption for assessment of prognosis. We analyzed the outcomes of 2231 patients with systolic heart failure who all underwent metabolic exercise stress testing on a modified Naughton protocol. During a mean follow-up of 5 years, 742 patients (33%) died and 249 patients (11%) underwent heart transplantation urgently. We found that, after accounting for baseline clinical characteristics, treadmill exercise time performed similar to peak oxygen consumption for predicting poor outcome. For a one minute change in exercise time there was a 7% increased hazard of death (e.g. comparing 480 to 540 seconds HR 1.07, 95% CI 1.02–1.12, P=0.004. Mortality rates were particularly high for women who exercised less than 5 minutes and 17 seconds and for men who exercised less than 6 minutes. Our findings suggest that treadmill exercise time may be valuable as an initial prognostic screening tool in patients with advanced heart failure.