Article Text
PDF
Abstract
Objective Resting pulse pressure (PP) is a risk factor for heart failure (HF); however, whether PP augmentation during exercise, a parameter easily obtained from routine treadmill stress testing, is associated with incident HF is unknown. Thus, we aimed to study the relationship between a novel parameter, the pulse pressure stress index (P2SI), and adverse outcomes among adults undergoing clinical exercise stress testing in the Henry Ford Exercise Testing Project.
Methods The P2SI was calculated as PP at peak exercise divided by resting PP and was analysed continuously and categorically using quartiles. Cox models examined the association between P2SI and adjusted HR (aHR) of incident HF, myocardial infarction (MI) or death. Receiver operating curve (ROC) analyses tested the optimal prognostic cut-point for P2SI.
Results Among 55 524 participants without prior MI or HF, mean (SD) age was 53 (13) years, 51% were men and 29% black. A total of 2516 HF, 1606 MI and 6224 mortality outcomes occurred. Quartile 3 P2SI (2.0–2.4) was chosen as the reference category based on ROC analyses. There was a graded inverse association of low P2SI with excess HF (aHR of 1.3 (95% CI 1.1 to 1.5) for quartile 2 and 1.5 (95% CI 1.2 to 1.8) for quartile 1, p for trend<0.001) and mortality (aHR of 1.1 (95% CI 1.01 to 1.2) for quartile 2 and 1.3 (95% CI 1.2 to 1.5) for quartile 1, p for trend<0.001). There was no association between P2SI and MI after adjustment. P2SI added significant prognostic information to more established stress testing parameters such as peak systolic blood pressure, per cent maximal predicted heart rate achieved and metabolic equivalents of task achieved.
Conclusions Poor augmentation of PP with exercise, specifically a P2SI below 2, is a novel and readily quantifiable exercise-based risk feature for HF and death.
- pulse pressure
- exercise
- heart failure
- death
- risk factor
Statistics from Altmetric.com
Introduction
Individuals with suspected coronary artery disease (CAD) are frequently referred for exercise stress testing. A drop in systolic blood pressure (SBP) with exercise, especially one associated with symptoms or ECG changes, is an indication for termination of the stress test because it is strongly suggestive of myocardial ischaemia.1 For this reason, it is standard practice to measure BP at regular intervals during exercise stress testing. As such, resting and peak exercise values of SBP, diastolic BP (DBP) and pulse pressure (PP) are routinely collected during stress tests obtained around the world, even in low resource settings where concomitant cardiac imaging is not available. However, the prognostic implications of changes in these BP parameters with exercise, in particular the significance of PP augmentation in response to exercise, remain relatively understudied despite the potential to use this information in the risk stratification of patients.2
Cardiac output (CO) and vascular compliance are major determinants of resting PP, which is a simply measured haemodynamic parameter that is prognostic for future adverse cardiovascular outcomes.3 With exercise, an increase in PP is driven by increase in CO and change in vascular compliance.4 Poor augmentation of PP with exercise may therefore be indicative of insufficient augmentation of CO (ie, left ventricular dysfunction) and/or poor compliance of the large arteries. In a 1958 study, the magnitude of PP rise with exercise stress testing was lower among persons with a prior history of anterior myocardial infarction (MI) than among controls with normal cardiac function.4 Given the above considerations, it stands to reason that exertional change in PP may be associated with risk for future heart failure (HF).5
To our knowledge, the only prior study to formally evaluate the prognostic implications of exertional change in PP found a 20% higher relative risk for mortality for every 10 mm Hg lower value in the difference between exercise PP and rest PP, confirming that poor augmentation of PP with exercise appears to be an adverse prognostic sign.2 However, these results did not account for resting PP and did not include data on non-fatal outcomes like HF. Using data from a large, real-world, clinical registry of adults referred for exercise treadmill stress testing, we studied the relationship between a novel exercise parameter—the pulse pressure stress index (P2SI)—and incident HF, MI and all-cause mortality. We hypothesised that P2SI is inversely associated with risk of these adverse outcomes.
Methods
Study cohort
The Henry Ford ExercIse Testing (FIT) Project is a cohort study that consists of 69 885 consecutive patients who underwent physician-referred treadmill stress testing at Henry Ford Health System-affiliated hospitals and ambulatory cares centres in Detroit, Michigan between 1991 and 2009. Details of the study are published elsewhere, including indications for the stress test referral (approximately 60% of participants had symptoms of chest pain or shortness of breath).6 The study cohort was limited to patients older than 18 years at the time of stress testing and excluded those undergoing modified or non-Bruce protocol stress tests.
Inclusion/exclusion criteria
We excluded individuals with a prior clinical history of HF (n=877), MI (n=10 190) or atrial fibrillation (n=1154). To minimise the influence of outliers and non-physiologic data points we also excluded participants below the 1st or above the 99th percentile of P2SI from the analysis (n=2140).
Exposure variable—P2SI
PP was calculated as the difference between SBP and DBP, respectively in millimetres of mercury. We defined a novel parameter, P2SI, as PP at peak exercise during stress testing divided by resting PP.
Outcome variables
Outcomes studied included all-cause mortality, MI and incident HF. Incident HF and MI were ascertained using appropriate International Classification of Diseases, Ninth Revision (ICD-9), codes and/or administrative claims files from services provided by Henry Ford or reimbursed by the patients' healthcare insurer. To avoid misclassifications, a new diagnosis of HF was made if ICD-9 code (428.XX) was documented on three separate encounters.6–10 Incident MI was also dependent on three separate encounters, using ICD-9 codes for MI. For non-fatal outcomes, FIT participants were administratively censored at their last contact with a provider in the Henry Ford Health System or when ongoing health coverage with the System’s health plan was no longer confirmed in order to minimise bias from loss to follow-up.
All-cause mortality was ascertained using an algorithm for searching the Social Security Death Index Death Master File that included social security number, first name, last name and date of birth. A complete algorithmic search was successful in over 99.5% of patients.6–10
In sensitivity analyses, we also tested a falsification end point, new diagnosis of diabetes mellitus (DM), which was identified when ICD-9 code for DM (250.XX) was listed on at least three separate encounters (for this sensitivity analysis we excluded persons with baseline DM, n=9930).
Covariates
Demographic information, indication for stress testing, anthropomorphic data, risk factor burden, past medical history and active medication use were obtained by nurses and/or exercise physiologists before each stress test. Resting heart rate (HR) and BP were manually assessed immediately prior to stress testing and also at peak exercise. For BP measurement, manual sphygmomanometers were used by trained nurses according to a defined protocol.6 Hypertension was defined as a self-reported prior diagnosis of hypertension, use of antihypertensive medications or electronic medical record database-verified diagnosis of hypertension. The BP at the time of the stress test was not used to classify participants in our analyses as hypertensive.
DM was defined as a self-reported previous diagnosis of DM, use of hypoglycaemic medications or a database-verified diagnosis of DM. Dyslipidemia was defined by a self-reported previous diagnosis of any major lipid abnormality, use of lipid-lowering medications or a database-verified diagnosis of dyslipidemia. Obesity was defined by self-report and/or assessment by the clinician. Current smoking was defined as self-reported active smoking at the time of the stress test. Exercise capacity, expressed in estimated metabolic equivalents of task (METs), was calculated by the Quinton treadmill controller based on peak speed and elevation. Maximum METs were categorised into four groups (<6, ≥6 to <10, ≥10 to 12 and ≥12).11 Exertional change in SBP and DBP was calculated as the difference between peak and resting BP values. Per cent HR achieved was calculated as HR at peak stress during exercise testing divided by maximal predicted HR (220–age).
A non-random subsample of FIT participants (n=13 134) had echocardiographic information on left ventricular ejection fraction (LVEF) available in the FIT dataset. This information was used in exploratory analyses testing the relationship between P2SI and LVEF.
Statistical analysis
Baseline characteristics were tabulated by P2SI quartiles and differences were tested using Χ2 tests for categorical variables and analysis of variance for continuous variables.
Cox proportional hazard models were used to study the association of P2SI and incident outcomes, modelling P2SI continuously (per 1 unit increase) and using quartiles. The proportionality assumption was verified using log-log plots. Model 1 adjusted for age, sex and race. Model 2 adjusted for model 1 covariates plus smoking, hypertension, diabetes, obesity, hyperlipidemia, antihypertensive medication use, statin medication use, aspirin, METs achieved during the stress test, resting HR, peak exercise HR, resting SBP, peak exercise SBP and resting PP. Interaction testing was performed between P2SI and age (<60 vs ≥60 years) and sex in model 2.
The association between continuous P2SI and incident outcomes was described graphically using unadjusted Kaplan-Meier curves and using multivariable adjusted restricted cubic splines centred at the 75th percentile with knots placed at the 10th, 50th and 90th percentiles of P2SI (adjusting for the variables in model 2).
To evaluate the ability of P2SI to discriminate incident events, we constructed receiver operating characteristic (ROC) curves. The optimal P2SI cut-off for prediction of incident outcomes (the point with best sensitivity and specificity) was then determined using the Liu method, which maximises the product of the sensitivity and specificity.12 To assess whether P2SI improved discrimination of incident outcomes, we calculated area under the ROC curves (the C-statistic) after adding P2SI to a base model and tested for significance using the likelihood ratio test. We also compared the potential incremental discriminatory ability of multiple exercise stress test variables (resting and peak HR, resting and peak SBP, METs achieved, per cent maximal HR achieved, resting PP, peak PP and P2SI) and outcomes of interest using area under ROC curves. The base model for this analysis was adjusted for demographics, traditional cardiovascular risk factors and medication use variables included in model 2.
Finally, we conducted the following sensitivity analyses: (1) we stratified our analysis by baseline age (<60 or ≥60 years), by stress test indication (chest pain/dyspnoea or not), by year of stress testing (before 2000 or after 2000, to assess for any temporal changes) and by baseline antihypertensive medication use (as these medications may directly influence PP measurements); (2) in models testing P2SI as a continuous exposure, we further adjusted for exertional change in SBP and exertional change in DBP; to determine whether they mediated any association between P2SI and outcomes and (3) we evaluated whether the association of P2SI and our primary outcomes could be due to residual confounding by using falsification testing with a ‘negative control’ outcome.13 14 For this falsification analysis, we chose the outcome of incident diabetes as we hypothesised that future diabetes would be unrelated to exertional change in either CO or vascular compliance (as reflected in P2SI) after adjustment for known cardiovascular risk factors.
A p value <0.05 was considered statistically significant. CIs were expressed as 95% CI. All analyses were performed using Stata/IC V.13.1 (StataCorp, College Station, Texas, USA).
Results
The baseline characteristics of the study cohort are displayed in table 1. Among 55 524 participants, mean (SD) age was 53 (13) years, 51% were men and 29% were black. Those in the uppermost P2SI quartile were on average younger, more likely to be men, black, obese, achieve higher METs, but were less likely to be hypertensive, dyslipidemic, diabetic and use cardiovascular preventive medications (all p<0.05).
Baseline characteristics of the study population overall and by pulse pressure stress index
Over a median follow-up time of 5.4 years (IQR 2.8–8.4), there were 2516 incident HF and 1606 incident MI events. There were 6224 mortality events over a median 10.3 years (IQR 7.6–14.3) follow-up. A progressive increase in unadjusted incidence rates of the three outcomes was observed with decreasing P2SI, with rates (per 1000 person-years) as high as 13.3 for HF, 7.7 for MI and 18.7 for death in the 1st quartile of P2SI (table 1).
In ROC analyses, the optimal P2SI cut-off for prediction of incident outcomes was 1.9 for all-cause mortality, 1.9 for MI and 1.8 for HF. We therefore chose quartile 3 of P2SI (2.0–2.4 mm Hg) as our reference category for the prediction of incident outcomes. Participants in the lower 2 quartiles of P2SI (values<2) had a significantly higher risk of HF and all-cause mortality (figure 1, table 2), while those in quartile 4 had a non-significantly lower risk of these outcomes. For example, quartile 1 was associated with a 47% higher adjusted risk of HF (95% CI 1.21 to 1.78, p<0.001) and 33% higher adjusted risk of all-cause mortality (95% CI 1.17 to 1.50, p<0.001). When analysed as a continuous exposure variable, P2SI was inversely and approximately linearly associated with both HF and death (figure 2). Results for MI were not significant after multivariable adjustment (table 2). Interaction testing between P2SI and age or sex was also not significant for all these outcomes.
Unadjusted Kaplan-Meier curves for the association of pulse pressure stress index (P2SI) with (A) incident heart failure and (B) all-cause mortality.
Multivariable adjusted linear and non-linear HRs (aHRs) (95% CI) for the association of pulse pressure stress index (P2SI) with (A) incident heart failure, and (B) all-cause mortality. Linear aHRs are presented in text within panels as per 1 unit increase in P2SI in linear Cox models. For the non-linear analysis, knots for restricted cubic splines are placed at the 10th, 50th and 90th percentiles of P2SI. The reference point for this continuous analysis was chosen as the upper bound of the reference third quartile group (ie, the 75th percentile). Model is adjusted for age, sex, race, smoking, hypertension, diabetes, obesity, hyperlipidemia, antihypertensive medication use, statin medication use, aspirin, metabolic equivalents of task, resting heart rate, peak heart rate, resting systolic blood pressure, peak systolic blood pressure and resting pulse pressure.
HRs (95% CI) for the association between P2SI and incident heart failure, myocardial infarction and all-cause mortality
Stratifying results by age, stress test indication, decade of enrolment and baseline antihypertensive medication use yielded similar results for all three outcomes (see online supplementary table 1). Further adjustment for exertional change in SBP or exertional change in DBP also yielded similar results, confirming that P2SI was independent of both (see online supplementary table 2). There was no significant association between P2SI and our negative-control outcome, incident DM, in any model (see online supplementary table 3).
Supplementary file 1
Using a simplified P2SI cut-point of <2 and adding it to the base model, the C-statistic changed from 0.802 to 0.807 for death (p<0.001), and for HF from 0.757 to 0.765 (p<0.001). The change in C-statistic after addition of other stress test parameter is displayed in table 3. Compared with P2SI, the only stress test parameters that consistently yielded a similar or higher C-statistics for HF and death were per cent HR achieved and number of METs achieved. When adding both per cent HR achieved and number of METs achieved to the variables included in the base model (expanded model), P2SI remained an independent predictor of both HF and death (table 3).
C-statistics for the addition of different stress test variables for discrimination of incident heart failure, myocardial infarction and all-cause mortality
In a post hoc exploratory analysis of 13 134 FIT participants with available LVEF information, the median LVEF in the first quartile of P2SI was 55% compared with 58% in the fourth quartile (p<0.001). Similarly, 422 of this subsample had an LVEF <40%, with approximately 50% of these individuals being in the first quartile of P2SI versus 12% in the fourth quartile (p<0.001).
Discussion
These data from a large cohort of patients referred for exercise stress testing support our original hypothesis that exertional change in PP during stress testing (P2SI) is inversely associated with HF and all-cause mortality; independent of traditional risk factors, exercise capacity/workload and change in other haemodynamic parameters (eg, METs achieved and HR and SBP at both rest and peak stress). There was no association with MI after adjustment.
With exercise, SBP is expected to increase. A decline in SBP below resting values15 and submaximal increase in SBP with exercise16 17 are known negative prognostic factors. An increase in DBP with exercise may reflect impaired vasodilation of vessels and has been associated with worse outcomes.18 An integrated approach that takes into account both SBP and DBP (ie, PP) may better reflect haemodynamic changes during exercise compared with either measurement alone. Indeed, our results show that exertional change in PP was inversely associated with HF and mortality independent of resting SBP, peak SBP and resting PP and also independent of exertional change in either SBP or DBP.
Our results add to a smaller study by Thomas et al 2 demonstrating excess mortality for every 10 mm Hg lower value in the absolute difference between exercise PP and rest PP. However, these authors did not have access to cause of death data nor did they adjust for baseline PP in their analysis (important because absolute change in PP from rest to peak exercise is influenced by baseline resting PP). Thus, it is challenging for the clinician to use this absolute PP difference as a memorable metric to gauge risk of HF and death in any given patient; as a 10 mm Hg difference for one patient with high resting PP will not have the same prognostic impact as the same difference in another patient with low resting PP.2 In our study, we derive P2SI as a novel stress parameter that is clinically intuitive, easy to calculate and one that can be applied to patients irrespective of their baseline PP.
Applying our findings to clinical practice, a P2SI<2 can be considered an easy to remember cut-point for HF risk and mortality risk that can be applied to any patient, irrespective of their baseline PP. The importance of adjusting for baseline resting PP in our analysis cannot be understated, as evident in table 1 where we found an inverse correlation between resting PP and P2SI (this was mathematically expected). One may ask how adults with low P2SI have increased risk for HF outcomes if low P2SI is also correlated with higher resting PP in our sample (because resting PP is known to be related to resting CO). We offer two potential explanations for this: 1) resting CO and change in CO with exertion may not be well correlated and 2) P2SI may be a better surrogate of poor vascular compliance in response to exertion. Indeed, because high resting PP also reflects poor vascular compliance (ie, increased afterload),19 this higher afterload at rest (ie, higher resting PP) may impair the ability of the vulnerable left ventricle to increase CO with exercise and achieve a higher P2SI. As such, a low P2SI may reflect mismatch between exercise-induced change in CO and vascular compliance. This possibility is consistent with our exploratory analysis of resting LVEF available in a subset of the patients studied. Further research examining the relationship between P2SI and change in LVEF with exercise as well as measures of rest and exertional vascular compliance (which are unfortunately not available in the FIT dataset) will be of interest.
Key strengths of our study include a large real-world clinical population of patients studied. P2SI is easy to measure, derived using simple haemodynamic variables and, importantly, could therefore be very useful in low resource settings where stress echocardiography and other forms of cardiac imaging are often unavailable. A low P2SI<2 during exercise testing might also trigger consideration for echocardiography referral to evaluate for LV dysfunction, particularly in settings where imaging access is limited—a hypothesis that warrants testing in confirmatory studies. Furthermore, persons with low P2SI should be considered for more aggressive risk factor control in order to reduce the risk of subsequent HF, particularly those with stage A disease.
However, our results should also be interpreted in the context of important limitations. Our study reflects the experience of a single health system and external generalisability needs to be tested in validation studies. On a technical basis, BP measurements were obtained by manual sphygmomanometer, which questions the reliability of DBP measures. However, our protocol was consistent with normal clinical practice and noise in the measurement of DBP would bias our results towards the null (indeed more precise measures of DBP would be expected to show stronger associations). Measurements of central PP were not performed in the FIT dataset; therefore, we could only examine the prognostic significance of peripheral pulse pressure. However, central PP is not routinely measured in clinical practice, so our results for peripheral PP are much more easily generalised and applied. Patients who completed a non-Bruce protocol for stress testing (eg, modified Bruce protocol or others) were not included in the FIT Project dataset, possibly eliminating subjects who are less fit. Potentially important stress test information including ECG changes, perfusion imaging, stress-induced echocardiographic results, maximum rate of oxygen consumption (VO2 max) and heart rate recovery were not available in the FIT dataset. We also did not have information on why stress tests were terminated and whether ischaemia was noted. Measurements of ejection fraction (EF) at the time of HF occurrence were not available in the FIT dataset either, nor were the ICD coding data suitable to accurately distinguish HF with preserved versus reduced EF. We therefore could not evaluate differences in the prognostic value of P2SI in relation to HF phenotypes.20 21 Similarly, as serial measurements of EF are not yet available in the FIT dataset, we could not study the association between pulse pressure and longitudinal change in EF among patients with HF.22 In any observational study, the possibility of residual confounding cannot be excluded (although results for our falsification outcome were consistently null).
In conclusion, a significant inverse relationship exists between exertional change in pulse pressure (P2SI) and risk of HF and all-cause mortality. Clinically, a P2SI<2 is a useful and simple threshold to identify that is associated with incident HF and mortality. As such, P2SI can be considered a novel and readily quantifiable high-risk feature derived from exercise stress testing that may be applied to any patient. Additional studies are required to further validate the clinical utility of this exercise parameter.
Key messages
What is already known on this subject?
Resting pulse pressure (PP) is associated with future heart failure (HF) and other adverse cardiac outcomes. Limited data suggest that PP augmentation with exercise may be an adverse prognostic sign; however, its relationship with specific outcomes including HF is unknown.
What might this study add?
PP augmentation with exercise added significant prognostic information for HF and death to more established stress testing parameters such as peak systolic blood pressure, per cent maximal predicted heart rate achieved and metabolic equivalents of task achieved.
How might this impact on clinical practice?
Poor PP augmentation with exercise, a parameter easily obtained from routine treadmill stress testing even in low resource settings without imaging, identifies persons at elevated risk for HF who may benefit from further screening and HF prevention interventions.
Acknowledgments
The FIT investigators acknowledge the critical input of patients who participated in this study.
References
Footnotes
Contributors Drafting: MA-R, JW. Statistical analysis: MA-R, JWM. Data acquisition: FR, CAB, SJK, JKE, MHA-M. Critical revision of the manuscript for important intellectual content: FR, CAB, SJK, JKE, MHA-M. Supervision: MJB, JWM.
Funding JWM is supported by a grant from the American Heart Association (17MCPRP33400031).
Competing interests None declared.
Patient consent Not required.
Ethics approval This study has been approved by the Henry Ford Health System Institutional Review Board.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement The data, analytic methods and study materials will not be made available to other researchers.