Cardiopulmonary exercise testing and prognosis in heart failure due to systolic left ventricular dysfunction: a validation study of the European Society of Cardiology Guidelines and Recommendations (2008) and further developments

Abstract

Aims

The study aims were to validate the cardiopulmonary exercise testing (CPET) parameters recommended by the European Society of Cardiology 2008 Guidelines for risk assessment in heart failure (HF) (ESC-predictors) and to verify the predictive role of 11 supplementary CPET (S-predictors) parameters.

Methods and results

We followed 749 HF patients for cardiovascular death and urgent heart transplantation for 3 years: 139 (19%) patients had cardiac events. ESC-predictors – peak oxygen consumption (VO₂), slope of minute ventilation vs carbon dioxide production (VE/VCO₂) and exertional oscillatory ventilation – were all related to outcome at univariate and multivariable analysis. The ESC/2008 prototype based on ESC-predictors presented a Harrell's C concordance index of 0.725, with a likely χ2 of 98.31. S-predictors – predicted peak VO₂, peak oxygen pulse, peak respiratory exchange ratio, peak circulatory power, peak VE/VCO₂, VE/VCO₂ slope normalized by peak VO₂, VO₂ efficiency slope, ventilatory anaerobic threshold detection, peak end-tidal CO₂ partial pressure, peak heart rate, and peak systolic arterial blood pressure (SBP) – were all linked to outcome at univariate analysis. When individually added to the ESC/2008 prototype, only peak SBP and peak O₂ pulse significantly improved the model discrimination ability: the ESC + peak SBP prototype had a Harrell's C index 0.750 and reached the highest likely χ2 (127.16, p < 0.0001).

Conclusions

We evaluated the longest list of CPET prognostic parameters yet studied in HF: ESC-predictors were independent predictors of cardiovascular events, and the ESC prototype showed a convincing predictive capacity, whereas none of 11 S-predictors enhanced the prognostic performance, except peak SBP.

Heart failure, prognosis, exercise testing

Introduction

In 2008, the European Society of Cardiology (ESC/2008) released the updated version of the Guidelines for the diagnosis and treatment of acute and chronic heart failure (HF),¹ in which three cardiopulmonary exercise testing (CPET) parameters are linked to poor prognosis: peak oxygen consumption (pVO₂), the slope of minute ventilation vs carbon dioxide production (VE/VCO₂ slope), and exertional oscillatory ventilation (EOV). In the interim, scientific research has shown several supplementary CPET parameters that are able to enhance outcome discrimination,^2–12 but, as usual in the case of an evolving and intriguing investigation area, more issues have been raised than clarified.¹³ Accordingly, the present study was conceived with two aims: first, to validate the ESC 2008 CPET (ESC-predictors) prognostic parameters in a large composite HF cohort, prospectively assembled and followed, and second, to verify the role of supplementary CPET parameters (S-predictors) for risk assessment.

Methods

Study population

We studied consecutive patients with chronic HF due to ischaemic disease or idiopathic dilated cardiomyopathy, referred to the Salvatori Maugeri Foundation, Veruno Scientific Institute for CPET as part of their functional evaluation. Eligibility criteria were: (1) echocardiographic left ventricular ejection fraction (LVEF) ≤40%; (2) ability to perform a symptom-limited CPET, with a peak respiratory exchange ratio (RER) ≥1.05; and (3) clinical stability defined as no change in NYHA class or absence of hospitalization for HF and stable medical treatment during the month before CPET. Medical treatment administered on the day of CPET was recorded.

Patients with primary valvular disease, myocardial infarction in the previous three months, severe peripheral vascular disease, chronic lung disease, neuromuscular disease or orthopaedic limitation were excluded.

Data were extracted from the Institution’s ergospirometry laboratory database, which are prospectively recorded and updated on a regular basis: both clinical characteristics and events at follow-up are revised on a yearly basis for each surviving patient, by means of a predetermined institutional protocol designed for internal quality-control activities. The database for this prospective research project was approved by the Institutional Review Board of the Salvatore Maugeri Foundation, and written informed consent was obtained from all patients prior to CPET.

Cardiopulmonary exercise testing

Symptom-limited CPET was performed on a bicycle ergometer with a ramp protocol of 10 W/min: details of the protocol have been published before.¹² Blood pressure was measured manually at rest and every 3 minutes during incremental exercise and at peak effort. The electrocardiogram and heart rate (HR) were monitored at rest and throughout exercise at 1-min intervals. Breath-by-breath respiratory gas exchange parameters were measured by a computerized metabolic cart (Sensormedics, Vmax29, Yorba Linda, CA).

ESC-predictors were evaluated for prognostic performance either separately or in combination, assembled creating an ESC prototype (ESC-prototype). ESC-predictors were determined as follows: (1) peak VO₂, expressed in ml/kg/min, as the mean value of VO₂ recorded during the previous 60 s of the exercise period in all patients, to improve its physiological validity in the presence of EOV);¹⁴ (2) VE/VCO₂ slope, calculated over the whole exercise period via least squares linear regression;¹⁵ (3) EOV, visually determined according to criteria from Kremser et al.¹⁶ as cyclic fluctuations in minute ventilation lasting more than 60% of the exercise duration and with an amplitude of more than 15% of the average amplitude of cyclic fluctuations.

Afterwards, 11 S-predictors were computed and their ability to improve the ESC-prototype’s predictive power was evaluated:

1. Predicted percentage of max VO₂ (ppVO₂) reached at peak effort, and estimated by a gender-, age-, height- and weight-adjusted and protocol-specific formula.¹⁷

2. Ventilatory anaerobic threshold (VAT) identification, detected using the V-slope method.¹⁸

3. Peak oxygen pulse (O₂pulse), delineated as peak VO₂ divided by peak HR and expressed in ml/beat.⁶

4. Peak RER, defined as the peak VCO₂/peak VO₂ ratio, as a 60-s averaged value.

5. Peak circulatory power (CP), calculated as the product of peak VO2 (ml/kg/min) and peak systolic blood pressure (mmHg).³

6. Peak VE/VCO₂ ratio (peak VE/VCO2), as mean value of the last 60 seconds of the exercise period.²

7. VE/VCO₂/VO₂, as VE/VCO₂ slope normalized by peak VO₂, with peak VO₂ expressed as l/min.⁵

8. Oxygen uptake efficiency slope (OUES), measured as the relationship between VO₂ and log₁₀VE.⁷

9. Peak end-tidal CO₂ partial pressure (PET-CO₂), recorded at peak exercise, as a 60-s averaged value.¹⁰

10. Peak HR, calculated as the mean of the last 60 s of peak effort.

11. Peak systolic arterial blood pressure (SBP), measured at the last exertional stage performed.

Echocardiography

Echocardiographic evaluations (Hewlett-Packard imaging system: model 77622-A) were performed within 5 ± 2 days of CPET in stable clinical and pharmacological conditions. Left ventricular ejection fraction (LVEF) was measured as described elsewhere.¹⁹

Follow-up and documentation of end-points

Patients were followed up at the outpatient clinic of our hospital and patients’ status was determined from the medical records. The follow-up of those who did not attend their scheduled appointments was obtained by telephone interview of the patient, patient’s family or primary care physician. Cardiac death and urgent heart transplantation (HT) were considered major events. Patients who did not experience a major event were followed up for a 3-year period. Clinicians conducting the CPET were not involved in decisions regarding cause of death or heart transplantation.

Statistics

Continuous data are expressed as mean ± SD. Student’s t-test for unpaired values was used to compare the means of groups for quantitative variables. For qualitative variables, the χ² test with Yates’ correction or Fisher’s exact test, if necessary, was employed. The level of statistical significance was set at a 2-tailed p value ≤0.05.

The prognostic value of each ESC-preditcor was studied using univariate Cox proportional-hazards regression;²⁰ a multivariable Cox proportional-hazards model comprising all ESC-predictors was then investigated (validation study). Although ESC-predictors have been endorsed one by one by the ESC 2008 guidelines, without a definition of an ESC prognostic model, we assembled all evidence-based CPET parameters in a so-called ‘ESC prototype’. Thereafter, the predictive ability of each of the 11 S-predictors was determined using univariate Cox proportional-hazards regression analysis: each 11 S-predictor that showed a significant association with the outcome (p < 0.01) was added to the ESC-prototype (ESC-prototype +) and Cox multivariable analysis performed. The likelihood test was then used to determine the difference between the ESC-prototype + and ESC-prototype (development study).

The following diagnostic and verification steps for Cox models were performed after each multivariable fit:

1. The proportional hazard assumptions of Cox models was verified by plotting smoothed scaled Schoenfeld residuals and by searching for the interaction between the covariate and time; linearity assumption for continuous variables was assessed graphically using martingale residuals;

2. Goodness of fit was analytically assessed using the Grønnesby and Borgan method²¹ and graphically with Cox-Snell residual plots.

3. Evaluation of the discrimination ability of the model was performed using Harrell's C concordance index (the proportion of all usable subject pairs in which the predictions and outcomes are concordant: a value of 0.5 indicates no predictive discrimination and a value of 1.0 indicates perfect separation of patients with different outcomes).

All calculations were performed using the STATA® 10 system (StataCorp, College Station, Texas USA).

Results

From July 1997 to July 2006, 749 HF patients were selected according to eligibility criteria. The study cohort was composed mostly of males (88%) with severe LV systolic dysfunction (mean LVEF 26 ± 7%) and in sinus rhythm (94%). Despite optimal medical treatment, including angiotensin-converting enzyme inhibitors (92%), diuretics (80%) and beta-blockers (65%), the majority of patients (n = 579, 77%) were in New York Heart Association (NYHA) functional class II to III, and mean exercise capacity was significantly impaired, reaching only 54% of ppVO₂, although mean peak RER was 1.2 (Tables 1 and 2).

Table 1.

Clinical characteristics of patients according to cardiac events

	Study cohort
Number of patients	749
Age (years)	59 ± 10
Male (%)	660 (88)
Ischaemic aetiology of HF (%)	529(71)
Atrial fibrillation (%)	48 (6)
NYHA class I (%)	170(23)
NYHA class II (%)	406 (54)
NYHA class III (%)	173 (23)
LV-EF (%)	26 ± 7
ARB (%)	63 (8)
ACE-I (%)	686 (92)
Loop diuretics (%)	600 (80)
Beta-blockers (%)	513 (69)

Data are expressed as mean value ± SD or number (%) of patients.

HF, heart failure; NYHA, New York Heart Association; LV-EF, left ventricular ejection fraction; ACE-I, angiotensin converting enzyme inhibitors; ARB, angiotensin receptor blockers.

Table 2.

Cardiopulmonary exercise testing parameters according to cardiac events

	Study cohort	Alive	Cardiac events	p value
Number of patients	749	610	139
Resting HR (beats/min.)	80 ± 14	80 ± 14	81 ± 15	0.573
Peak HR (beats/min.)	130 ± 19	131 ± 19	126 ± 19	0.009
Resting SBP (mmHg)	118 ± 17	119 ± 17	112 ± 17	0.000
Peak SBP (mmHg)	157 ± 26	160 ± 24	142 ± 26	0.000
N. VAT identification (%)	198(26)	141(23)	57(41)	0.000
Peak VO2 (l/min)	1.08 ± .44	1.14 ± .47	0.86 ± .28	0.000
Peak VO2 (ml/kg/min)	14.7 ± 4.0	15.3 ± 3.9	12.3 ± 3.2	0.000
% predicted peak VO2 (%)	54 ± 14	56 ± 14	44 ± 13	0.000
Peak CP (mmHg x ml/kg/min)	2341 ± 829	2471 ± 811	17731 ± 651	0.000
Peak VCO2 (l/min)	1.24 ± .40	1.29 ± .39	1.03 ± .35	0.000
Peak VCO2 (ml/kg/min)	17.0 ± 4.7	17.6 ± 4.7	14.5 ± 4.0	0.000
Peak RER	1.2 ± 0.1	1.2 ± 0.1	1.2 ± 0.1	0.005
Oxygen pulse (mL/beat)	8 ± 3	9 ± 3	7 ± 2	0.000
Peak VE (l/min)	45 ± 12	46 ± 12	41 ± 12	0.000
VEVCO2 slope	36 ± 7	35 ± 7	39 ± 7	0.000
Peak VE/VCO2	37.4 ± 7.1	36.5 ± 6.7	41.1 ± 7.7	0.000
VE/VCO2/VO2	2.6 ± 1.1	2.503 ± 1.0	3.50 ± 1.4	0.000
OUES (l/min)	1436 ± 432	1492 ± 419	1192 ± 401	0.000
Peak PET-CO2 (mmHg)	33 ± 5	33 ± 5	31 ± 5	0.000
Exertional oscillatory ventilation (%)	126 (17)	92(15)	36 (25)	0.003

HR, heart rate; SBP, systolic blood pressure; VAT, ventilatory anaerobic threshold; VO₂, oxygen consumption; VCO₂, carbon dioxide production; CP, circulatory power; RER, respiratory exchange ratio; VE, minute ventilation; VE/VCO₂ slope, slope of regression relating VE to VCO₂; VE/VCO₂, ratio between VE and VCO₂; VE/VCO₂/VO₂, VE/VCO₂ slope normalised by peak VO2; OUES, oxygen uptake efficiency slope; PET-CO₂, partial pressure of end-tidal CO₂.

During the follow- up, 139 (19%) patients had major cardiac events. Actuarial 1-and 3-year survival rates were 96% and 84%, respectively. Mean follow-up for patients with and without events was 404 ± 275 and 3 years, respectively. Peak VO₂, and VE/VCO₂ slope were significantly lower and higher, respectively, in non-survivors compared to survivors, while EOV was more frequently detected in non-survivors (Table 2). All ESC-predictors were significantly associated with outcome both at univariate and at multivariable Cox hazards analysis (Tables 3 and 4): the Harrell's C concordance index for the ESC-prototype was 0.725, with a likelihood χ² of 98.31.

Table 3.

Relationship of ESC-predictors and S-predictors to outcome at univariate Cox regression analysis

	HR	95% CI	p value	χ2
ESC-predictors
Peak VO2	0.81	0.77–0.85	.000	63.26
VE/VCO2 slope	1.06	1.04–1.08	.000	45.21
EOV	1.86	1.28–2.73	.001	10.41
S-predictors
Peak HR	0.98	0.98–0.99	.008	7.00
Peak SBPa	0.97	0.96–0.98	.000	61.95
% Predicted peak VO2	0.93	0.92–0.95	.000	82.11
VAT	2.13	1.52–2.98	.000	19.25
Oxygen pulse	0.73	0.68–0.80	.000	54.74
Peak RER	8.10	1.95–33.64	.004	8.30
Peak CP	0.99	0.99–0.99	.000	81.33
Peak VE/VCO2	1.06	1.05–1.08	.000	51.33
VE/VCO2/VO2	1.56	1.43–1.71	.000	92.70
OUES	0.998	0.99–0.99	.000	56.37
Peak PET-CO2	0.91	0.88–0.94	.000	34.01

CPET, cardiopulmonary exercise testing; HR, hazard ratio. CI, confidence intervals. For other abbreviations, see Table 2.

a

Per 5 mmHg increase.

Table 4.

ESC-predictors related to outcome at multivariable Cox regression analysis

	HR	95% CI	p value	χ2
Peak VO2	0.83	0.78–0.87	.000	44.75
VE/VCO2 slope	1.03	1.01–1.05	.004	8.31
EOV	2.37	1.61–3.49	.000	19.46

All S-predictors were linked to outcome at univariate Cox hazards analysis (Table 3). When individually added to the ESC-prototype (Table 5), only peak SBP and peak O₂ pulse significantly improved the model discrimination ability. The ESC-prototype + with peak SPB had a Harrell's C index of 0.7503, and reached the highest likelihood, χ² = 127.16 (p < 0.0001). Peak RER, OUES, peak PET-CO₂, peak HR, VE/VCO₂/VO₂, VAT identification and peak VE/VCO₂ did not provide additional prognostic information, whereas ppVO₂ and peak CP were not tested because of co-linearity (Table 5). Importantly, peak VO₂, VE/VCO₂ slope, EOV and peak SBP (ESC-prototype + SBP) maintained their independent predictive value when tested at multivariable Cox hazards analysis (Table 6).

Table 5.

Difference in prognostic performance between the ESC-prototype and ESC- prototypes +

CPET - MODELS	LR χ2	p	Harrel C
ESC-prototype	98.32		0.7257
ESC + peak SBPa	127.16	.000	0.7503
ESC + oxygen pulse	107.00	.013	0.7314
ESC + peak RER	102.84	NS	–
ESC + OUES	102.25	NS	–
ESC + peak PetCO2	99.98	NS	–
ESC + peak HR	98.74	NS	–
ESC + VE/VCO2/VO2	98.67	NS	–
ESC + VAT	98.35	NS	–
ESC + peak VE/VCO2	98.32	NS	–
ESC + % predicted peak VO2	–	b	–
ESC + peak CP	–	b

ESC, European Society of Cardiology model, which embraces three ESC evidenced-base parameters: peak VO₂, VE/VCO₂ and EOV (ESC prototype). ESC +, European Society of Cardiology integrated prototype (ESC + models are created by adding individually all univariate significant S-predictors to ESC prototype). NS, not significant. Other abbreviations see Tables 2 and 3.

a

Per 5 mmHg increase.

b

Not tested due to co-linearity.

Table 6.

ESC + prototype with peak SBP at multivariable Cox regression analysis

	HR	95% CI	p value	χ2
Peak VO2	0.85	0.81–0.907	.000	29.88
VE/VCO2 slope	1.02	1.00–1.04	.029	4.79
EOV	2.44	1.66–3.59	.000	20.82
Peak SBPa	0.906	0.87–0.940	.000	27.77

Abbreviations see Tables 2 and 3.

a

Per 5 mmHg increase.

Discussion

Healthcare is full of innovations, but they are slow to reach clinical practice, if they arrive there at all. This is the case in the setting of CPET and risk assessment in HF: numerous risk CPET parameters have been proposed, but few have been adopted in clinical practice. The present study was conceived to refine the list of CPET risk parameters developed through ongoing research, evaluating 14 ‘predictive’ CPET parameters, by means of two investigative routes: validation of the ESC-predictors in a sizeable HF cohort, and an analysis of its predictive power with other CPET parameters added to the ESC prototype, created assembling the ESC-predictors (ESC-prototype). Thus, our results can be summarized under two main headings: (1) validation of the ESC-prototype’s prognostic performance and (2) implementation of an ESC-prototype (ESC-prototype +).

Validation of the ESC-prototype’s prognostic performance

Although it is not the largest HF cohort undergoing symptom-limited CPET assembled for risk stratification purposes, the present collection is unique in that it includes the highest number of patients with EOV as well as VE/VCO₂ slope determination. Our results confirm, revise and extend to 3 years the predictive performance of ESC-predictors – peak VO₂, VE/VCO₂ slope and EOV – investigated in smaller HF series that evaluated these parameters separately or in comparative binary manner.^15,22–28 All ESC-predictors were highly predictive and, peculiarly, EOV showed the highest risk profile: thus EOV is confirmed as a robust ominous ventilatory pattern in HF, irrespective of different definition criteria.²⁹

In addition, for the first time, we documented that ESC-prototype, created assembling the ESC-predictors, is an efficient predictive paradigm: this finding confirms that HF is a disease of holistic exercise physiology, involving unfavourable interactions among muscles, vessels and the heart, and an intricate interplay between sympathetic and parasympathetic efferent activity and mechanical and chemical reflexes.^30,31 Thus, risk definition according to exercise capacity cannot be described by a single CPET parameter. Indeed, peak VO₂ depends on the efficacy of VE, which includes lung diffusion, cardiac output, oxygen extraction by the muscles and muscle function, whereas the VE-related parameters (VE/VCO₂ slope and EOV) depend on the ventilatory efficiency, which is related to reflex activity, dead space ventilation and, possibly, the amount of gas to be exchanged. In truth, VO₂- and VE-related parameters portray different but interrelated aspects of exercise adaptations in HF¹³ and no CPET parameter looks at a single body function alone. Consequently, the debate about the predictive superiority of VO₂- vs VE-related parameters in HF is obsolete, while the usefulness of a multi-parametric CPET interpretation for risk discrimination should be stressed.

Implementation of an ESC-prototype+ (development study)

In the last decades, an exciting and informative literature has witnessed the proposal of new S-predictors for prognostic prediction.^2–12,32 In this study we have investigated 11 S-predictors, each of which has persuasive credentials in favour of its clinical application: ppVO₂, which takes into account age, gender and weight, provides a more accurate aerobic capacity impairment definition;³² both VAT and peak RER are reliable descriptors of maximal effort, and are related to anaerobic metabolism activation and exercise-induced metabolic acidosis;⁴ O₂ pulse is a non-invasive indicator of stroke volume and arteriovenous O₂ difference, more dependent on cardiac pump function reserve than peak VO₂ because it integrates HR response;⁶ peak SBP and CP are equally key parameters for the description of the ‘hydraulic’ responses of the heart and the circulatory system during exercise;^5,14 peak VE/VCO₂ is easy to calculate;^2,22 both VE/VCO₂/VO₂ and OUES improve the definition of increased VE response to exercise, connecting VE and VO₂ adaptation during exercise;^5,7,33 and PET-CO₂, either measured at VAT or at peak exercise, is strongly related to HF severity and cardiac output.³⁴ Alone or in combination, all these sound CPET parameters have provided robust prognostic information in the different small HF cohorts^2–12,32 in which their outcome discrimination ability was validated. Interestingly, we found that none of the 11 S-predictors distorted the ESC-prototype’s prognostic performance, and most of them did not improve its predictive ability, except peak SBP and peak O₂ pulse, with peak SBP accomplishing the highest statistical effect.

Two studies have screened several CPET parameters for outcome assessment in HF due to LV dysfunction.^12,34 Myers et al.³⁵ studied 5 CPET risk parameters, peak VO₂, VE/VCO₂ slope, OUES, resting PET-CO₂ and HR recovery, defined as maximal HR minus HR at 1 min in recovery, in 710 HF patients who were followed for cardiovascular death (primary end-point) and composite end-points (HT, left ventricular assist device implantation, and hospitalisation for HF) for a mean 29 ± 25 months. A multivariate 0–15 score was developed, including all five CPET parameters expressed dichotomously, determined with the receiver operating characteristic (ROC) curve analysis: each dichotomous CPET parameter was weighted according to the hazard ratios for cardiovascular death and composite events and summed to calculate the composite score. A significant step-wise increase in both mortality and composite outcome rates associated with increasing weighted summed scores was evident, and the estimated 1-year death rates were 26.6% with summed scores >15 and 0.4% with scores <5. The summed risk score was a more accurate predictor of outcomes than any individual CPET variable.

Recently, we examined 12 CPET parameters:¹² peak VO₂, ppVO₂, VAT detection, VO₂ at VAT, O₂ pulse, peak CP, VE/VCO₂ slope, peak VE/VCO₂, OUES, EOV, peak SBP and peak HR, in 631 HF patients, all treated with carvedilol, and followed for cardiovascular death over 3.8 ± 1.4 years. EOV, although poorly represented, was associated with higher total mortality rate compared to non-EOV (43 vs 11%), while peak SBP and peak CP were the most predictive of prognosis in HF patients without EOV, more than peak VO₂, or VE/VCO₂ slope. At multivariate analysis, peak CP disappeared because of co-linearity with VO₂ and blood pressure, and peak exercise SBP emerged as the highest prognostic parameter.

The concept that lower peak SBP conveys prognostic information in HF is not new.³⁶ Although, peak SBP is not a perfect surrogate of cardiac output, its prognostic impact is thought to be linked with the ‘hydraulic’ responses of the heart and the circulatory system. Physiologically, SBP reflects the performance of the cardiac pump in the context of its function, having the ability to generate both flow and pressure: in the complex interplay of circulatory readjustment during exercise, the increase of arterial pressure is the result of multiple stimulatory effects of the mass discharge of the sympathetic nervous system throughout the body, including vasoconstriction of the arterioles in most body tissue beside the active muscles, increasing pumping activity of the heart and increase in mean systemic filling pressure.³⁷ Thus, since heart contribution to arterial pressure is fundamental, and low pressure-generating capacity is a hallmark of poor outlook³⁸ it is not surprising that CPET parameters that mirror the pumping capability and reserve of the heart, such as peak SBP, may become reference indexes in the assessment of HF patients for prognostic screening.³⁹

The present study is unique in terms of its innovative method: we did not create any new predictive CPET-prototype, but launched the ESC-prototype without any CPET cut-point values, to support the notion that each CPET-parameter expresses a genuine continuum of risk (except for EOV). We suggest a ESC-prototype +, bearing in mind two extremes: patients with high VE/VCO₂ slope, low peak VO₂ and SBP and with EOV are exposed to the highest hazard, and those with low VE/VCO₂ slope, high peak VO₂ and SBP, in absence of EOV are at low risk. The remaining patients with residual combination of trend of CPET variables show an intermediate risk definition.

From the present findings the need emerges for at least two future investigations. First, since no prediction model should be implemented in practice until its performance has been validated, the ESC-prototype + (with peak SBP) prognostic performance should be tested in a different HF population, prospectively collected, using wider inclusion criteria. Second, although sophisticated risk adjustment techniques allow outcome data to be compared among different types of patients, much variability remains unexplained. Indeed, as a general rule, a clinical trial with a broad-spectrum population has less relevance for a given complex patient; but specific study populations will not necessarily be more relevant for a given patient whose complexities differ from those of the population studied. Regrettably, the present study does not solve this problem of generalizability, since younger (<50 years) patients, females and those with atrial fibrillation were largely under-represented. Hence, our findings also need to be confirmed for the ‘unconventional’ CPET HF patients.

Limitations

Our study has some limitations. Although we have evaluated the longest list of CPET parameters so far studied, it is inevitable that the list is incomplete. More sophisticated and fascinating CPET indexes are now available⁴⁰ but we believe that model simplicity and measurement reliability are important criteria in developing clinically useful prognostic tools. Moreover, this is a retrospective analysis; nevertheless, three criteria were taken into account to minimise analysis contamination. First, data were obtained from a database that is prospectively updated on a regular basis, according a predetermined protocol. Second, CPET execution and supervision (symptom-limited testing and peak RER >1.05), CPET variables calculation (i.e. VE/VCO₂ slope, EOV, OUES and so on) and follow-up management were all carried out on an independent basis, and data were assembled in blind fashion (by Fabio Comazzi). Third, HF patients who had been enrolled in a recent previous trial¹² were discarded to prevent the possibility of reproduction and misrepresentation of results.

Conclusions

The ESC-prototype, including peak VO₂, VE/VCO₂ slope and EOV, is efficiently predictive in a large HF cohort, and its prognostic performance is not altered by the addition of other supplementary CPET parameters, except peak SBP. Thus, we recommend an amplified model (ESC-prototype +) based on four key CPET-parameters: three ESC-predictors plus peak SBP. Importantly, our findings support a general shift in approach from a medical logic based on the judgment of a single CPET risk factor, to one centred on the combination of more CPET risk parameters to describe overall risk in HF. Further studies are warranted to confirm these results.

Acknowledgments

The authors are grateful to Alfio Agazzone, Elena Bonanomi and Barbara Temporelli for technical support, to Fabio Comazzi for data management and statistical analysis, and to Rosemary Allpress for her careful revision of the English manuscript.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Conflict of interest

The authors declare they have no conflicts of interest.