Two-minute heart rate recovery after cycle ergometer exercise and all-cause mortality in middle-aged men

Authors


Kai P. Savonen, Kuopio Research Institute of Exercise Medicine, Haapaniementie 16, 70100 Kuopio, Finland.
(fax: 358 17 2884488; e-mail: savonen@student.uef.fi) and

Rainer Rauramaa, Kuopio Research Institute of Exercise Medicine, Haapaniementie 16, 70100 Kuopio, Finland.
(fax: 358 17 2884488; e-mail: rainer.rauramaa@uef.fi).

Abstract

Abstract.  Savonen KP, Kiviniemi V, Laaksonen DE, Lakka TA, Laukkanen JA, Tuomainen T-P, Rauramaa R (Kuopio Research Institute of Exercise Medicine, Kuopio, Finland; Kuopio University Hospital, Kuopio; Information Technology Center, University of Eastern Finland, Kuopio Campus, Kuopio; Kuopio University Hospital, Kuopio; Institute of Biomedicine/Physiology, University of Eastern Finland, Kuopio Campus; Institute of Public Health and Clinical Nutrition, University of Eastern Finland, Kuopio Campus; and Lapland Central Hospital, Rovaniemi, Finland). Two-minute heart rate recovery after cycle ergometer exercise and all-cause mortality in middle-aged men. J Intern Med 2011; doi: 10.1111/j.1365-2796.2011.02434.x

Background.  A slow heart rate recovery (HRR) after an exercise test is associated with an increased risk of all-cause mortality in asymptomatic individuals, but the data regarding additional prognostic information provided by HRR beyond other exercise test variables are inconsistent. We investigated the prognostic significance of HRR for premature death, particularly in relation to other exercise test variables.

Methods.  The study subjects were a representative population-based sample of 1102 men (42–61 years of age) without cardiovascular disease, cancer or diabetes. HRR was defined as the difference between maximal HR and HR 2 min after a maximal symptom-limited exercise test using a cycle ergometer. The association between HRR and premature mortality was examined with Cox regression models.

Results.  During an average follow-up of 18 years, 238 deaths occurred. HRR was an independent predictor of death [for a decrease of 12 beats min−1, relative risk (RR) 1.16, 95% CI 1.02–1.33, = 0.02] after adjustment for age and established risk factors. When added in a Cox model with chronotropic response (decrease of 12 beats min−1, RR 1.09, 95% CI 0.93–1.27, = 0.26) or cardiorespiratory fitness (decrease of 12 beats min−1, RR 1.12, 95% CI 0.98–1.30, = 0.08), the association between a slow HRR and an increased risk of death was clearly weaker.

Conclusion.  A slow 2-min HRR after a cycle ergometer exercise test was an independent predictor of death in healthy middle-aged men after accounting for demographic and clinical characteristics. However, it was no longer predictive after accounting for chronotropic response and exercise capacity.

Introduction

A slow heart rate recovery (HRR) after an exercise test is associated with an increased risk of all-cause death in individuals without cardiovascular disease (CVD) [1–5]. However, whether HRR provides additional prognostic information beyond other exercise test variables is debated. A slow HRR has been shown to predict premature mortality independent of cardiorespiratory fitness (CRF), chronotropic incompetence (ChI) or ST-segment depression in electrocardiography (ECG) during exercise testing [1–5]. By contrast, the predictive value of HRR is not apparent after adjustment for peak heart rate (HRpeak) [6] or heart rate (HR) reserve [7]. Additionally, HRR did not improve the prognostic power of a model that already contained CRF, ST-segment depression during exercise testing, HRpeak and systolic blood pressure (SBP) at submaximal work during exercise testing [8]. The purpose of this study was to investigate whether HRR provides additional prognostic information for the prediction of the risk of all-cause mortality in a population-based sample of middle-aged men free of CVD at baseline and to examine whether other exercise test variables dilute the predictive value of HRR.

Methods

Subjects

This study included participants of the Kuopio Ischaemic Heart Disease Risk Factor Study (KIHD), an ongoing population study designed to investigate risk factors for CVD and related outcomes in men from East Finland [9]. Study population constituted a representative sample of men who lived in the town of Kuopio or in neighbouring rural communities at baseline between March 1984 and December 1989. Of 3235 eligible men, 2682 (82.9%) participated in the study.

Heart rate recovery values at 2 min after the exercise test were available for 1929 men. We excluded men with CVD (including coronary heart disease, cardiac insufficiency, cardiomyopathy, claudication and stroke; n = 615), cancer (n = 35) and diabetes (n = 118) and those who used beta-blockers (n = 334). After these exclusions, the final study population included 1102 men. The study protocol was approved by the Research Ethics Committee of the University of Kuopio and complied with the Declaration of Helsinki. Each participant provided written informed consent.

Assessment of HRR and other exercise test variables

A maximal, symptom-limited exercise test using an electrically braked cycle ergometer was performed at baseline by 2361 men [10]. HR was recorded from an ECG at rest, at the end of each 60-s interval during the exercise test, at peak exercise and during recovery. During recovery, the workload was set to 0 watts and subjects were allowed to continue pedalling at a self-chosen frequency if desired. No predefined pedalling frequency was used during recovery, in contrast to the previous study in which work intensity during recovery was fixed [3]. HRR was defined a priori as the reduction in HR from HRpeak to HR at 2 min after the exercise test to maximize the number of subjects included in the analyses, because values of HR at 1 min after the exercise test were only available for 1068 men. HR increase from 40% to 100% of maximal work capacity in the exercise test (HR40-100) was calculated to characterize ChI [11]. CRF was quantified as peak oxygen uptake (VO2peak), which was defined as the highest value recorded over a 30-s interval. The presence of ischaemic changes in the ECG during the exercise test, defined as ST-segment depression of >1.0 mm at 80 ms after the J point, was considered an abnormal exercise test ECG.

Assessment of other risk factors

The collection of blood samples and the assessment of cigarette smoking, alcohol consumption, blood biochemistry, body mass index (BMI) and SBP at rest have been described elsewhere [10, 12]. Serum C-reactive protein (CRP) was measured with an immunometric assay (Immulite High Sensitivity C-reactive protein assay, Los Angeles, CA, USA).

Determination of deaths

Deaths were ascertained by computer linkage to the National Death Registry using social security numbers. There were no losses to follow-up. All deaths that occurred between study enrolment (from 20 March 1984 to 5 December 1989) and 31 December 2007 were included in the analyses. The average time to death or the end of follow-up was 18.8 years (range 0.8–23.8 years). In this study sample of 1102 men, 238 (21.6%) deaths occurred during the follow-up period.

Statistical analysis

Differences in baseline characteristics between men with HRR above and below the median HRR were tested with age-adjusted linear and logistic regression analyses as well as the Mann–Whitney U-test. The skewed distribution of serum CRP was corrected via logarithmic transformation in all analyses.

Kaplan–Meier survival curves were constructed and age-adjusted, and multivariable Cox proportional hazards regression models were used to study the association between HRR and the risk of all-cause death. First, HRR was forced into the multivariable Cox model with potentially important predictors of death, including age, alcohol consumption, BMI, cigarette smoking, plasma fibrinogen, serum low-density lipoprotein (LDL) cholesterol, resting SBP and serum CRP. In the next phase, we examined whether HRR improves the performance of survival models already including exercise-induced myocardial ischaemia, CRF or ChI at a time.

The change in global goodness-of-fit and discrimination ability as well as reclassification of risk was used to explore whether HRR brings any additional information into models in which it is not included. The change in global goodness-of-fit was calculated using the likelihood ratio (LR) chi-squared statistic with the modified Hosmer–Lemeshow chi-squared statistic [13]. For the latter, small values indicate a good fit of the model, whereas values exceeding 20 indicate significant lack of fit (< 0.01). The discrimination ability of each model was assessed by the calculation of an optimism-corrected c index for censored data based on 200 cycles of bootstrap resampling [14]. The bootstrap method was used for comparison of c indices between models. We also evaluated the increased discriminative value of HRR by calculating the net reclassification improvement (NRI) and the integrated discrimination improvement (IDI) using a previously described method [15]. These statistics characterize the difference between two models (in this study, the multivariable model with and without HRR) with regard to the individual estimated probability that a case subject will be categorized as such. An increased probability that case subjects will be categorized as case subjects and a decreased probability that control subjects will be categorized as case subjects imply better prediction ability, whereas the opposite implies worse prediction ability. For NRI, we used an arbitrarily chosen cut-off value, which categorizes 10% of subjects into the high-risk group. In the current cohort, the cut-off value corresponded to predicted 68% survival probability at 18 years of follow-up. IDI reflects the reclassification of subjects in a similar manner to NRI, but it considers the change in the estimated prediction probability as a continuous variable.

In supplementary multivariable survival analyses, HRR at 2 min after the cessation of exercise was replaced by HRR at 1 min in men for whom the information was available. In the current study, a total of 312 otherwise eligible men were excluded because of the missing HRR values. To explore the potential selection bias, differences in baseline nonexercise test characteristics between the excluded and included men were tested with age-adjusted linear and logistic regression analyses as well as the Mann–Whitney U test. Age-adjusted Cox proportional hazards regression model was used to compare mortality between the excluded and included men.

Assessments of the modified Hosmer–Lemeshow chi-squared statistic, c index, NRI and IDI were based on predicted and actual 18-year survival rates. The Cox proportional hazards assumption was confirmed by the inspection of the scaled Schoenfeld residuals. In all analyses, < 0.05 was considered statistically significant. Statistical analyses were performed using spss 14.0 for Windows (SPSS, Inc., Chicago, IL, USA) and R software version 2.6.1 (Vienna, Austria). The Design and HMisc libraries of Harrell were used for model validation.

Results

The age of the subjects ranged from 42 to 61 years, and the mean (SD) HRR was 41 (12) beats min−1. The overall risk profile was more unfavourable in men with HRR below the median value of 40 beats min−1 (Table 1); the only exception was exercise-induced myocardial ischaemia, which was more prevalent in men with a faster HRR.

Table 1.   Characteristics of the study cohort at baseline
Clinical characteristicMean (SD), median (range) or percentage
All men (n = 1102)HRR ≤40 beats min−1 (median value), n = 567HRR >40 beats min−1 (median value), n = 535P-value for difference between groupsa
  1. HRR, heart rate recovery.

  2. aDifference between groups of HRR at 2 min either below or above the median value. Differences in categorized variables were tested with logistic regression analysis and in continuous variables with linear regression analysis after adjustment for age. However, differences in age and cigarette smoking were tested with the Mann–Whitney U test. bCigarette-years denotes the lifelong exposure to smoking, which was estimated as the product of years of smoking and the number of cigarettes smoked daily at the time of examination [12]. cThe criteria for myocardial ischaemia during the exercise test were ischaemic changes in ECG defined as ST-segment depression of >1.0 mm at 80 ms after the J point.

 Number of deaths238 (21.6%)155 (27.3%)83 (15.5%)<0.001
 Age (years)51 (42–61)52 (42–61)51 (42–61)<0.001
 Men with alcohol consumption ≥94 g week−1 (highest quartile)25.0%27.7%22.1%0.01
 Body mass index (kg m−2)26.4 (3.3)26.7 (3.4)26.1 (3.0)0.001
 Cigarette smoking (cigarette-years)b0 (0–2880)0 (0–2880)0 (0–2700)<0.001
 Serum C-reactive protein (mmol L−1)1.10 (0.10–53.50)1.28 (0.10–53.50)0.91 (0.10–29.30)<0.001
 Serum LDL cholesterol (mmol L−1)3.85 (0.93)3.90 (0.96)3.80 (0.90)0.12
 Plasma fibrinogen (g L−1)2.95 (0.52)3.03 (0.52)2.86 (0.51)<0.001
 Systolic blood pressure at rest (mmHg)132 (15)134 (16)131 (14)0.01
Exercise test characteristics
 HRR at 2 min (beats min−1)41 (12)32 (7)50 (9)<0.001
 Heart rate increase from 40% to 100% of maximal work (beats min−1)56 (13)51 (13)62 (11)<0.001
 Peak oxygen uptake (L min−1)2.60 (0.61)2.47 (0.57)2.75 (0.62)<0.001
 Myocardial ischaemia during exercisec7.0%5.5%8.6%0.04

HRR, nonexercise test risk factors and mortality

Men with HRR below of 40 beats min−1 had greater unadjusted mortality than men with a faster HRR (Table 1, Fig. 1). In the age-adjusted Cox model, the risk of death increased by 35% (95% CI 18–56, < 0.001) for a 1-SD (12 beats min−1) decrement in HRR. After adjustment for other potentially important predictors, the risk of death increased by 16% (2–33, = 0.024) for a 1-SD decrement in HRR (Table 2). The global goodness-of-fit improved when HRR was introduced to the model as evaluated by a change in the LR chi-squared statistic from 155.1 to 160.3 (= 0.022 for increase). The modified Hosmer–Lemeshow chi-squared statistic for the model without and with HRR was 9.6 (> 0.2) and 5.7 (> 0.6), respectively, with both values suggesting an excellent fit. As a marker reflecting discrimination ability, the optimism-corrected c index was 0.72 in both models (i.e. without and with HRR). When HRR was expressed as a dichotomized variable, men with HRR below the median value of 40 beats min−1 had a 1.37 (95% CI 1.04–1.79, = 0.025) times higher risk of death than those with higher HRR values.

Figure 1.

 Kaplan–Meier survival curves for men with heart rate recovery of ≤40 or >40 beats min−1.

Table 2.   Risk factors for death in study cohort: multivariable modela
Risk factorRelative risk (95% CI)P-value
  1. CI, confidence interval.

  2. aFor continuous variables (except age and cigarette smoking), the relative risks were calculated for a change of 1 SD, as shown. For serum C-reactive protein (CRP), a logarithmically transformed value was used in survival analysis and a 1 SD change is shown as an antilogarithm.

Age (per 1-year increment)1.07 (1.05–1.10)<0.001
Alcohol consumption ≥94 g week−1 (highest quartile vs. others)1.38 (1.03–1.85)0.03
Body mass index (per increment of 3.3 kg m−2)1.01 (0.88–1.16)0.88
Cigarette smoking (per increment of 100 cigarette-years)1.08 (1.04–1.11)<0.001
Serum CRP (per increment of 2.59 mmol L−1)1.22 (1.05–1.42)0.01
Serum LDL cholesterol (per increment of 0.93 mmol L−1)1.16 (1.02–1.31)0.02
Plasma fibrinogen (per increment of 0.52 g L−1)1.14 (0.99–1.31)0.08
Systolic blood pressure at rest (per increment of 15 mmHg)1.25 (1.11–1.40)<0.001
Heart rate recovery at 2 min (per decrement of 12 beats min−1)1.16 (1.02–1.33)0.02

A total of 190 subjects died during the first 18 years of follow-up. Reclassification for subjects who died and for those who did not die during the first 18 years of follow-up is summarized in Table 3. For seven subjects who died, reclassification was more accurate when the model with HRR was used, and it was less accurate for any subjects who died. Among the subjects who did not die, six were reclassified in a lower-risk category and seven were reclassified in a higher-risk category. After the addition of HRR, the estimated NRI was 0.036 (= 0.01) and the estimated IDI was 0.007 on an absolute scale (= 0.005) or 5.6% as a relative increase.

Table 3.   Reclassification of subjects who died or who did not die during the first 18 years of follow-up
Predicted 18-year risk of death with established risk factorsaPredicted 18-year risk of death with established risk factors and HRRa
<32%b≥32%bTotal number of subjects
  1. HRR, heart rate recovery.

  2. aEstablished risk factors include all those in Table 2 except HRR. bPredicted 18-year risk of death with established risk factors and HRR. cPredicted 18-year risk of death with established risk factors.

Subjects who died
 <32%c1347141
 ≥32%c04949
 Total number of subjects13456190
Subjects who did not die
 <32%c8437850
 ≥32%c65662
 Total number of subjects84963912

HRR, other exercise test variables and mortality

In multivariable Cox models, exercise-induced myocardial ischaemia was not associated with risk of death, whereas a low CRF and ChI were both strong predictors of death (Table 4). When HRR was introduced to models including other exercise test variables in turn, the magnitude of the associations between other exercise test variables and mortality did not change considerably (Table 4). With exercise-induced myocardial ischaemia, HRR remained an independent predictor of death but the association between HRR and mortality weakened when CRF and particularly ChI were included in the model (Table 4). Of various combinations of two exercise test variables in the same model, by far the most predictive of mortality (based on the LR chi-squared statistic) was the combination including CRF and ChI. In that model, both a low CRF (= 0.03) and ChI (= 0.04) were independent predictors of death. The LR chi-squared statistic of the model was 167.9.

Table 4.   The effect of introduction of HRR into the models with other exercise test variablesa
Exercise-induced myocardial ischaemia and established risk factors in modelExercise-induced myocardial ischaemia, established risk factors and HRR in model
Relative risk (95% CI) of death for myocardial ischaemiaP-valueRelative risk (95% CI) of death for myocardial ischaemiaP-valueRelative risk 95% CI) of death per decrement of 1 SD in HRRP-value
  1. CI, confidence interval; SD, standard deviation; HRR, heart rate recovery; VO2peak, peak oxygen uptake; HR40-100, heart rate increase from 40% to 100% of maximal work.

  2. aEstablished risk factors include all those in Table 2 except HRR.

1.19 (0.72−1.96)0.501.24 (0.75–2.05)0.401.18 (1.02–1.35)0.02
Peak oxygen uptake and established risk factors in model Peak oxygen uptake, established risk factors and HRR in model
Relative risk (95% CI) of death per decrement of 1 SD in VO2peakP-valueRelative risk (95% CI) of death per decrement of 1 SD in VO2peakP-valueRelative risk (95% CI) of death per decrement of 1 SD in HRRP-value
1.27 (1.09–1.49)0.0031.23 (1.05–1.45)0.011.12 (0.98–1.30)0.08
Chronotropic incompetence and established risk factors in model Chronotropic incompetence, established risk factors and HRR in model
Relative risk (95% CI) of death per decrement of 1 SD in HR40-100P-valueRelative risk (95% CI) of death per decrement of 1 SD in HR40-100P-valueRelative risk (95% CI) of death per decrement of 1 SD in HRRP-value
1.23 (1.08–1.43)0.0041.19 (1.01–1.39)0.041.09 (0.93–1.27)0.26

Heart rate recovery brought additive predictive information to a multivariable survival model including exercise-induced myocardial ischaemia, when assessed by the LR chi-squared statistic, but the changes in optimism-corrected c index and reclassification of risk were not statistically significant (Table 5). When introduced into the survival model including VO2peak as a measure of CRF, no change was observed in optimism-corrected c index and only a trend was seen (= 0.08 for change in LR chi-squared statistic) for improvement in fit of the model with HRR. From the viewpoint of risk reclassification, however, HRR provided additional discrimination to the model including VO2peak as evidenced by a statistically significant value for IDI. HRR did not improve the performance of the multivariable survival model that included HR40-100 as a measure of ChI regardless of performance measure used.

Table 5.   Assessment of performance of survival models after introduction of HRR into the modela
 Change in likelihood ratio chi-squared when HRR is included in modelP-valuec index without/with HRR in modelP-value for differenceIntegrated discrimination improvement with HRRP-value
  1. HRR, heart rate recovery; HR40-100, heart rate increase from 40% to 100% of maximal work.

  2. aEstablished risk factors include all those in Table 2 except HRR.

Exercise-induced myocardial ischaemia and established risk factors in model5.460.020.72/0.720.450.00330.16
Peak oxygen uptake and established risk factors in model3.100.080.72/0.720.560.00480.01
HR40-100 and established risk factors in model1.270.260.72/0.720.500.00150.22

HRR at 1 min and mortality

Among 1068 men in whom data regarding HRR at 1 min were available, 230 (21.5%) deaths were observed during follow-up. In multivariable Cox model without other exercise test variables, the risk of death increased by 10% (−4–27, = 0.15) for a 1-SD (10 beats min−1) decrement in HRR at 1 min. The association between HRR at 1 min and mortality weakened (= 0.46) when HR40-100 was included in the model, whereas the association between HR40-100 and mortality was still statistically significant (= 0.03).

Comparison between the excluded and included men

When the nonexercise test variables shown in Table 1 were considered, the excluded men were older (53 vs. 51 years, < 0.001) and had a higher mean LDL cholesterol level (4.36 vs. 3.85 mmol L−1, < 0.001) than the included men. The age-adjusted relative risk of death was 1.00 (95% CI 0.79–1.28, = 0.98) in the excluded men, compared with those included in the analysis.

Discussion

The main finding of the present study was that a slow HRR after a maximal, symptom-limited exercise test using a cycle ergometer was an independent risk factor for death in a population-based sample of middle-aged men free of CVD, cancer or diabetes after accounting for demographic and clinical characteristics. However, a slow HRR did not predict death after taking exercise capacity and chronotropic response into account. Accordingly, the obtained prognostic information from the exercise test was optimized by considering both exercise capacity and chronotropic response along with established risk factors, but HRR did not improve risk prediction.

The present findings are in line with the results of the previous studies in which a slow HRR was shown to be an independent risk factor for premature death in individuals without CVD [1–5]. However, the present study adds to the current knowledge, because we critically evaluated the prognostic value of HRR by addressing specifically the issue of whether HRR brings additional predictive information beyond other exercise test variables. None of the previous studies [1–5] included such strict methodological considerations [16] as in the current study.

In the present study, the association between HRR at 1 min after the exercise test and mortality was statistically nonsignificant. Nonetheless, in absolute terms, the magnitude of the association between HRR 1 min after the exercise test and mortality was not very different from that of HRR at 2 min after the exercise test. Only in one study [6] was the prognostic value of HRR considered both at 1 and 2 min after the exercise test; in this case, a similar prognostic yield for both HRR measures was found.

Exercise-induced ST-segment depression indicating myocardial ischaemia did not predict death in the present study, but men with HRR below the median value had a higher prevalence of exercise-induced ischaemic ECG changes than those with a higher HRR. These findings suggest that neither a slow HRR itself nor its association with an increased mortality is explained by exercise-induced myocardial ischaemia.

Heart rate recovery improved the performance of the multivariable Cox model including CRF as measured by the reclassification of risk, but the improvement was only close to statistical significance when assessed by change in the LR chi-squared statistics. Both HRR and CRF were shown to be independent risk factors for death when entered into the same model [2–4], and the results of the current study are in line with these findings. These observations suggest that the association between a slow HRR and an increased risk of death is not explained by a low CRF although a direct association between HRR and CRF was observed. Indeed, previous reports [2, 17] have shown that a low CRF and a slow HRR potentiate the effect of each other as risk factors for premature death.

Of importance, the prognostic value of HRR weakened clearly when HR40-100 as a marker of ChI was introduced into the model. This observation is in line with the results of two previous studies in asymptomatic individuals [6, 7]. Furthermore, adding HRR into the model of risk of coronary death that already included HRpeak improved the predictive value of the model only slightly [8]. Only one large follow-up study [1] reported that HRR was an independent predictor of death after controlling for ChI. The present results agree with most of the available data [6–8], suggesting that HRR may not bring additional prognostic information beyond ChI in asymptomatic subjects. In a study by Desai et al. [18], the difference in HRR between patient groups was markedly reduced after correction for chronotropic response, and the authors concluded that measuring HRR adds little additional information to chronotropic response.

The indices characterizing the reclassification of risk (NRI and IDI) have been shown to be sensitive enough to reveal statistically significant improvement in the reclassification with a new marker in situations in which the c index has shown no improvement [15]. The reclassification indices have been criticized for being too sensitive to show improvements in risk prediction as statistically significant yet clinically providing only limited additional prognostic information [19]. In the present study, too, HRR improved the prognostic power of the model including VO2peak when assessed with the IDI, whereas the improvement was not observed using the overall goodness-of-fit and the c index. It is not possible to be certain whether an increase of 5.6% in the reclassification of risk is clinically important as it depends on viewpoint (e.g. cost-effectiveness considerations). Further studies are needed to critically evaluate the current methods for the assessment of utility of new risk markers.

We did not use a predefined work intensity during cool down or passive rest immediately after the test, in contrast to the previous studies in which work intensity during recovery has been fixed [3] or the subjects have been put to a sitting or supine position immediately after the test [1, 2, 4, 6–8]. This discrepancy explains a lower median HRR at 2 min after the exercise test in the present study (40 beats min−1) than in two previous studies that included a passive cool down (49 and 50 beats min−1) [1, 6]. At the time of collection of the KIHD baseline data during the period 1984–1989, the prognostic value of HRR was unknown and no attention was paid to the standardization of the cool-down period. The nonstandardized cool-down period probably results in an underestimation of the risk associated with a slow HRR, because the increased random variation gives rise to a decreased signal-to-noise-ratio, thus compromising the accuracy of HRR in prediction models.

A strength of the present study is that the participation rate was high with no losses to follow-up. In addition, we have reliable data on mortality, because deaths were ascertained using the National Death Registry with social security numbers. Furthermore, the comprehensive assessment of risk factors for death allowed us to investigate the independent association between HRR and premature mortality.

A limitation of the present study is that only middle-aged men were enroled. Subjects who used beta-blockers were excluded from the current study, but the use of beta-blockers has been an exclusion criterion in most previous studies of asymptomatic men [1, 4, 6–8]. Finally, we do not know whether HRR changed during the long follow-up period and how possible changes could have affected our results.

Conflict of interest statement

No conflict of interest was declared.

Acknowledgements

We thank the staff of the Kuopio Research Institute of Exercise Medicine and the Institute of Public Health and Clinical Nutrition, University of Eastern Finland, for helping with data collection.

Ancillary